An Elaborate Tension Receptor System Highlights Sensory Complexity in the Hind Leg of the Locust

The neural circuitry that controls movements of animal limbs generally contains both central driving elements and peripheral sensors providing feedback to the system. Insect legs are driven by motor neurones contained within each segmental ganglion. The actions of these motor neurones are coordinated by groups of local or intersegmental interneurones, some of which must comprise the rhythm-generating circuits that underlie repetitive movements such as walking. Each leg contains many proprioceptors that respond to different mechanical aspects of its activity and which feed back onto the motor neurones and interneurones controlling it. For example, the position, velocity and acceleration of the femoro-tibial joint are registered by the femoral chordotonal organ (locust, Zill, 1985; Matheson, 1990, 1992a,b; stick insect, Hofmann et al. 1985; Hofmann and Koch, 1985). A single strand receptor associated with the metathoracic femoral chordotonal organ responds to tibial extension (Bräunig, 1985). Campaniform sensilla on the proximal tibia monitor cuticular distortions that result from resisted contractions of the extensor tibiae muscle (Burrows and Pflüger, 1988), and superficially placed multipolar receptors (Coillot and Boistel, 1968, 1969; Williamson and Burns, 1978; Bässler, 1977) respond to movements of the femoro-tibial joint. The prothoracic and mesothoracic legs of the locust contain another multipolar receptor, which is sensitive to tension generated by the main flexor tibiae muscle. It may act as a sensor in a load compensation circuit (Theophilidis and Burns, 1979).

The metathoracic leg of the locust is specialised for kicking and jumping (e.g. Heitler, 1974; Bennet-Clark, 1975), although it is also used for walking. Burns (1974) suggested that this specialisation of function was a reason why the metathoracic femoral chordotonal organ apparently contained only one of the two groups of homologous sensilla present in the other legs. Furthermore, the flexor tension receptor present in the prothoracic and mesothoracic legs has not been observed in the metathoracic leg (Theophilidis and Burns, 1979). We have recently shown that the metathoracic femoral chordotonal organ does possess both groups of sensilla present in the other legs (Matheson and Field, 1990). In the present paper, we describe a multipolar receptor neurone that responds to tension generated by the accessory flexor tibiae muscle of the metathoracic leg. This receptor therefore appears to be homologous to that found in the flexor tibiae muscle of the other legs, providing further evidence that the functional specialisation of the metathoracic legs has not been accompanied by a marked reduction in their sensory supply.

We show that the accessory flexor tibiae muscle, which is absent in the prothoracic and mesothoracic legs (Heitler, 1974; Theophilidis and Burns, 1979), is innervated by only two of the nine excitatory flexor motor neurones, by two common inhibitory motor neurones and, probably, by one dorsal unpaired median (DUM) neurone. At least seven other flexor tibiae motor neurones do not innervate the accessory flexor muscle. This means that the animal has partially independent control over the amount of force generated in the two flexor muscles. The receptor responds only to the force generated by the few fibres of the accessory flexor and not to force generated by the ‘main’ flexor muscle. The tension receptor is somewhat similar to arthropod muscle receptor organs (MROs). For example, the MRO at the coxa–trochanter joint of the locust leg consists of a single multipolar receptor neurone associated with a group of three muscle fibres that lie parallel to, but are innervated independently of, the trochanteral levator muscle (Bräunig and Hustert, 1985a; Bräunig et al. 1986).

The partially independent control of the accessory flexor muscle suggests that, of the nine known excitatory flexor tibiae motor neurones, those that innervate the accessory flexor muscle could act somewhat similarly to β-motor neurones of vertebrate muscle spindles, which innervate both extrafusal and intrafusal muscle fibres. The muscle spindles of amphibians, like the locust accessory flexor muscle, receive only such shared (β) motor innervation.

Adult locusts were taken from crowded cultures in Cambridge, England (Schistocerca gregaria Forskål), or Christchurch, New Zealand (Locusta migratoria L.). All electrophysiological experiments were performed on S. gregaria. Morphological observations were carried out on both species.

The tension receptor was stained through its axon in n5B1 using conventional backfilling techniques (5 % cobalt chloride, 3–7 days at 4 °C). The preparation was fixed in Carnoy’s fixative, cleared in methyl salicylate and viewed using an interference contrast microscope (Leitz).

For electron microscopy, tissue was fixed in 2.5 % glutaraldehyde buffered with 0.1 mol l^-1 sodium cacodylate (pH 7.4), for 3 h at 4 °C. It was then rinsed in the buffer for 1 h, post-fixed in 1 % osmium tetroxide (buffered as above) for 3 h, rinsed again (all at 4 °C), and then dehydrated in an ethanol series at room temperature. The dehydrated tissue was transferred to propylene oxide, embedded in Spurr’s epoxy resin and sectioned on an LKB 8800 ultrotome. Sections were stained in 3 % uranyl acetate followed by lead citrate, and viewed on a Jeol 1200E × transmission electron microscope (TEM).

Animals were fixed ventral side up in Plasticine. The left hind leg was held at right angles to the body and rotated posteriorly so that its anterior surface faced upwards. Nerves and muscles were exposed where necessary by cutting small windows in the anterior femoral cuticle. Sensory and motor activity was recorded from appropriate nerves using 50μm diameter silver wire hook electrodes. Muscle activity was recorded using 50 μm diameter silver wires inserted through the cuticle into the relevant muscle. Nerves and muscles could be stimulated by passing current through the recording electrodes.

Forces produced by nearly isometric contractions of the tibial muscles were measured using pairs of foil strain gauges mounted on stiff steel probes. Each probe was calibrated with a series of known masses spanning the range of forces to be measured. The relationship between voltage and force for each probe was linear in the range used. To measure the overall force produced by all the muscles acting at the femoro-tibial joint, a probe was waxed directly onto the tibia of the leg. To measure the force produced by the accessory flexor muscle alone, a pair of custom-made forceps was attached to a probe so that it could grasp the flexor apodeme (on which the accessory flexor inserts). The apodeme was then cut proximal and distal to the forceps, leaving only fibres of the accessory flexor muscle attached. To produce controlled movements of the muscle apodeme, the probe was attached to a Ling vibrator controlled by a custom-built ramp generator (Matheson and Ditz, 1991). The relationship between the linear position of the apodeme and the femoro-tibial angle was determined in five animals (not shown, see also Field and Burrows, 1982). This relationship allowed us to express all apodeme movements as equivalent changes in the femoro-tibial angle.

The mesothoracic and metathoracic ganglia were exposed by cutting a window in the ventral thoracic cuticle. Some overlying air sacs were removed, and the ganglia were supported on a wax-coated silver platform. To facilitate penetration by microelectrodes, the sheath of the metathoracic ganglion was treated with a 1 % (w/v) solution of protease (Sigma type XIV). Electrodes had resistances of approximately 60 MO when filled with 1 mol l^-1 potassium acetate.

Recordings were made from the somata of identified tibial motor neurones and midline spiking interneurones in the metathoracic ganglion. The extensor muscle is innervated by two excitatory motor neurones, a common inhibitor motor neurone (CI₁) and a dorsal unpaired medial neurone (DUMETi) (Hoyle, 1978). The flexor muscle is innervated by two common inhibitors (CI₂, CI₃; Hale and Burrows, 1985) and at least nine excitatory motor neurones (Phillips, 1981) divided into three groups: anterior, posterior and lateral. Each group contains a fast, a slow and an intermediate neurone (Hoyle and Burrows, 1973).

The motor neurones were identified by the positions of their somata and by the following additional features. The fast extensor tibiae motor neurone (FETi) was identified by its antidromic spike in response to electrical stimulation of the extensor muscle. The slow extensor (SETi) was identified by a 1:1 correlation of its spikes with potentials recorded in the extensor muscle. Flexor tibiae motor neurones were identified by their characteristic depolarisations following FETi spikes (Burrows et al. 1989). The type of flexor motor neurone (fast, intermediate or slow) was determined by simultaneously or sequentially recording from all three flexors in the particular group (see Parker, 1993). CI₁ was identified by a 1:1 correlation of its spikes with potentials recorded in the extensor nerve. CI₂ and CI₃ were identified by 1:1 correlations between their spikes and potentials recorded in the flexor nerve (n5B2).

Patterns of motor innervation were investigated by (a) intracellular stimulation of flexor tibiae motor neurones in the metathoracic ganglion, and (b) stimulation of n5B2 with gradually increasing voltage while recording intracellularly from accessory flexor muscle fibres.

To detect the central connections made by the tension receptor, its spikes recorded from n5B1 in mid femur triggered a signal averager that was used to detect postsynaptic potentials in central neurones. A simultaneous recording of the tension receptor spike in n5 near the ganglion was used to calculate both the afferent conduction velocity and, knowing the relevant distances between recording sites, the unexplained central delay (i.e. the period left unaccounted for by the estimated time taken for the afferent spike to travel from n5 to the intracellular recording site).

Throughout this paper, values are presented as mean ± S.E.M. unless stated otherwise.

The tension receptor (TR) is a large multiterminal sensory neurone (type II; Zawarzin, 1912) with a mean soma diameter of 34 μm (range 26–41 μm, N=5). It attaches near the proximal origin of the anterior accessory flexor muscle (136c), with its dendrites oriented towards the origin of the muscle fibres on the cuticle (Fig. 1A). It has 1–4 primary dendrites, which range in length from 97 to 240 μm (N=16 dendrites in seven neurones) and generally lack secondary branches (Fig. 1D). They intermingle with fibres of the accessory flexor muscle (Fig. 1C), but do not obviously extend as far as the origins of the muscle fibres on the cuticle. The soma gives rise to an axon which, together with an axon from a hair sensillum on the anterior dorsal femur, forms a tiny short nerve extending directly towards the junction of n5B1 and n5B1e, with which it merges.

The accessory flexor muscle comprises two isolated distal bundles of fibres inserting onto the anterior and posterior surfaces of the flexor tibiae muscle apodeme. Because these are slightly separated from the main flexor fibres (M136a) and have more dorsal origins on the anterior and posterior surfaces of the femur, they were designated by Snodgrass (1929) as separate muscles (M136b,c). Only the anterior accessory flexor (M136c, Fig. 2A) is associated with a multipolar receptor neurone. This muscle, which contains 13 fibres (N=5), is innervated at four points by a distal femoral branch of n5B2 (and not n5B1, which carries the tension receptor axon), which also has branches into several distal fibres of the main flexor (Fig. 2A,B). Electron micrographs of transverse sections of the accessory flexor motor branches showed a maximum of five axons (Fig. 2C), although most of the branches amongst (e.g. arrowheads in Fig. 2B) or apposed to the muscle fibres (Fig. 2E) had only three axons. The neuromuscular synapses (not shown) contained electron-lucent synaptic vesicles.

The muscle fibres are characterised by relatively sparse sarcoplasmic reticulum and a high thin:thick filament ratio of 7:1 (Fig. 2E,F; cf. Hoyle, 1978). The sarcoplasmic reticulum was usually associated with transverse tubules as dyads, but it did not segregate the contractile proteins into discrete bundles (‘fibrils’, Elder, 1975) typical of insect phasic muscle. Sarcomere length was 8.7±0.6 μm (S.D., N=191), which is typical of insect slow tonic muscle (Elder, 1975; Hoyle, 1975).

Hook electrodes placed on sensory nerve 5B1 in mid-femur record the activity of many afferent neurones in the leg, including those of the femoral chordotonal organ. When all branches of n5B1 distal to the accessory flexor muscle, including that to the femoral chordotonal organ, were ablated, a single tonic spike remained. Cutting n5B1 proximal to the accessory flexor muscle or crushing the proximal ends of the muscle fibres (where the tension receptor cell body lies, see Fig. 1) abolished this tonic activity. We therefore conclude that the tonic spikes in a single axon come from the sensory neurone in the accessory flexor muscle.

With the femoro-tibial angle set at approximately 90 °, the spike frequency of the receptor in quiescent animals ranged from 6 to 25 Hz (12.5±1.6 Hz, N=14). The firing rate was uniform, with few bursts or doublets of spikes. The conduction velocity, measured between hook electrodes in mid-femur and near to the metathoracic ganglion, ranged from 0.9 to 1.5 m s^-1 (1.1±0.03 m s^-1, N=27).

To determine whether the tension receptor responds primarily to isometric tension or to movement, the accessory flexor muscle was activated in two ways. First, it was induced to contract actively by tickling the animal with a soft paintbrush or by injecting current into particular flexor motor neurones so that they spiked. Second, it was passively stretched by extending the tibia or pulling on the flexor apodeme.

Voluntary isometric contractions caused the TR firing rate to increase (Fig. 3Ai,ii, Bi). More powerful flexions were accompanied by larger changes in firing frequency (Fig. 3Ai,ii), up to a maximum recorded of 175 Hz. High frequencies of receptor firing were maintained throughout periods of isometric contraction lasting several seconds. When the motor nerves to the accessory flexor muscle were ablated, leaving the motor innervation to the main flexor muscle intact, active isometric flexions were not accompanied by increases in the TR firing rate (compare Fig. 3Bi and Bii), indicating that it does not respond to the tension generated by the main flexor muscle. Under the same conditions, active isometric extensions also had no effect on the TR firing rate (Fig. 3Biii).

In contrast to the above, passive extension and flexion movements had little or no effect on the tonic firing rate of the TR (Fig. 3Aiii,iv, C). The tonic firing frequency was lowest when the tibia was nearly fully flexed (10 °), increased by 53 % (from 11±2 Hz to 17±3 Hz, N=4) when the tibia was fully extended (160 °) and, on return to 10 °, showed little hysteresis. A similar increase was seen in two preparations in which the motor innervation to the accessory flexor muscle had been cut, although the firing rates in these preparations were 12 % lower than those for the intact preparations. The maximum tonic firing rate measured during passive stretch was therefore only 11 % that of the maximum rate recorded during active isometric contractions.

The relationship between passive tension in the accessory flexor muscle and TR firing rate was examined in more detail by isolating the accessory flexor muscle from the main flexor muscle. During slow (66 ° s^-1) stretch and release movements from the equivalent of 80 ° to 160 °, the TR firing rate closely paralleled the changes in passive force (Fig. 3Di). During faster (267 ° s^-1) stretches, the TR responded with a marked, but very transient, increase in firing rate (Fig. 3Dii). During these faster movements, however, the forceps tended to vibrate on the springy strain gauge. Experiments without the strain gauge yielded similar results, indicating that vibrations did not contribute to the observed response (Fig. 3Ei–iii). In fact, the lack of response to the damped oscillation (e.g. Fig. 3Dii) shows that the tension receptor is relatively insensitive to imposed vibration. Instantaneous firing frequency reached 90 Hz during a movement at 533 ° s^-1.

Two slow excitatory flexor motor neurones (those from the anterior and lateral groups) innervate the accessory flexor muscle. They also innervate the main flexor muscle. In some, but not all, recordings they were shown to receive input from the tension receptor (see below). The accessory flexor tibiae muscle is innervated by both of the common inhibitory motor neurones that also innervate the main flexor muscle (CI₂, CI₃). They receive excitatory inputs from the tension receptor (see below).

Stimulation of the motor nerve (not shown) suggests that different accessory flexor muscle fibres are innervated by different numbers of motor neurones. The following combinations were recorded: two excitors and one inhibitor; two excitors and two inhibitors; two excitors; one excitor. The most frequent pattern (two excitors and one inhibitor) corresponds with the commonest innervation pattern (three axons) observed in TEM sections of the accessory flexor muscle (Fig. 2D).

During supramaximal stimulation of the motor nerve (n5B2), tension recordings of the semi-isolated accessory flexor muscle showed contraction profiles characteristic of slow muscle. For stimulation frequencies in the range 10–50 Hz, tension slowly increased (e.g. 0.02 g per 2 s at 10 Hz) in proportion to stimulus frequency without reaching a plateau (Fig. 4A–D); the minute individual twitches showed marked facilitation (200 % after 20 twitches, Fig. 5A). Furthermore, single twitches had a slow relaxation phase. The same stimulus conditions delivered to the main flexor muscle gave rapid twitch contractions, which reached a plateau within five twitches (Fig. 4F,G) and showed barely detectable facilitation (18 % after 20 twitches, Fig. 4E). Single twitches relaxed to baseline in 33–50 % of the observed relaxation time of single accessory flexor twitches. These results corroborate the finding that slow motor neurones innervate the accessory flexor muscle.

The slow flexor motor neurone from the anterior group, which innervated the accessory flexor muscle, fired rapidly during struggling, and the tension receptor firing rate increased (Fig. 5A). Leg movement was blocked. The change in TR firing was only apparent at the higher frequencies of motor neurone activity, suggesting that a threshold level of muscle activation had to be exceeded before the receptor was affected. Depolarising this flexor motor neurone to increase its firing frequency also caused an increase in the TR firing rate (Fig. 5B). Hyperpolarisation of the flexor motor neurone to suppress its tonic firing did not reduce the TR firing rate below the resting level.

The relationship between the firing frequency of another flexor motor neurone (controlled by injection of current) and the TR firing rate is shown in Fig. 5C. At normal resting potential this motor neurone fired at approximately 26 Hz. In this case also, there was a threshold rate of motor neurone activity (between 10 and 16.5 Hz) below which no further alteration in TR firing rate occurred. The tension receptor did not generally respond markedly to individual motor spikes. In one case, however, each excitatory muscle potential elicited a marked transient increase in the TR firing rate (Fig. 5E).

Intracellular recordings from the accessory flexor muscle showed that, when the muscle received a barrage of hyperpolarising potentials from one of the common inhibitor motor neurones, the TR firing rate was reduced (Fig. 5D, arrow), although the overall flexor force did not change. Individual inhibitory inputs from CI₂ had little or no apparent effect on the TR firing rate (Fig. 5E). In Fig. 5D, the initial burst of large spikes recorded in the main flexor myogram was coincident with a marked increase in total force, but this was not reflected in either the muscle fibre recording or the tension receptor firing rate. The discrepancy between the pattern of tibial force and the pattern of tension receptor firing (which should reflect accessory flexor tension) indicates that the accessory flexor can, indeed, be activated at least partially independently of the main flexor. We cannot rule out the further possibility that the receptor responds to the force generated by only some of the 13 muscle fibres of the accessory flexor muscle. The wide spread of its dendrites suggests that it contacts most of the fibres.

All three common inhibitory neurones received depolarising inputs from the tension receptor (Fig. 6A–C), with central latencies of 0.9–1.8 ms (mean 1.4±0.1 ms, N=8; see Fig. 8C). The input to CI₁ was clear in only two of six recordings from this cell.

The extensor tibiae muscle of the locust hind leg is innervated by one slow motor neurone (SETi) and one fast motor neurone (FETi). Tension receptor spikes had no effect on FETi, but elicited hyperpolarising potentials in SETi, with a central delay of 3.7–8.1 ms (Figs 7A, 8C). It is unlikely, therefore, that this effect was monosynaptic. The inhibitory nature of the PSP is confirmed in Fig. 9A, which shows that an increase in TR firing frequency was followed by a reduction in SETi firing frequency. Background synaptic inputs to SETi made it impossible to determine whether every TR spike was followed by an IPSP, but signal averaging always revealed the same inhibitory effect (N>12).

In quiescent animals, the tension receptor had excitatory effects in 17 recordings of flexor motor neurones (Fig. 7B), inhibitory effects in 5 (Fig. 7C), no effects in 40, and unclear effects in 3 (N=65). A depolarising input from the TR was most commonly recorded in slow (42 %), less often in intermediate (23 %) and least often in fast (12 %) flexor motor neurones. There was no evidence for a different occurrence of TR depolarising input to flexor motor neurones in the three different anatomical groups (anterior, 30 %; posterior, 25 %; lateral, 27 %). The central delay for the depolarising potentials ranged from 2.2 to 3.8 ms (mean 3.0±0.1 ms, N=16), suggesting a polysynaptic pathway.

In active animals, these depolarising inputs were sometimes replaced by longer-latency hyperpolarising or mixed inputs (6–10 ms; mean 8.0±0.6 ms, N=6; Figs 7D, 8C). This difference in timing between the initial inflections of depolarising and hyperpolarising potentials is evident in Fig. 7D.

Five recordings of spiking interneurones with somata near the ventral midline of the metathoracic ganglion revealed depolarising inputs from the tension receptor (Fig. 8A). Another neurone, which could not be induced to spike with depolarising current, and whose soma was near to, but clearly not that of, CI₃, also received a depolarising input (Fig. 8B). The central latencies of these inputs ranged from 1.5 to 2.5 ms (mean 2.0±0.2 ms, N=6; see Fig. 8C).

Electrically stimulating the accessory flexor muscle caused the muscle to contract and the tension receptor firing rate to increase (Fig. 9). This caused the tonic firing of SETi to slow down or stop (Fig. 9A). In a different preparation, similar electrical stimulation led to a burst of large depolarising potentials in an anterior flexor tibiae motor neurone (Fig. 9B). The potentials in the flexor motor neurone did not match the pattern of spikes in the tension receptor, indicating the contribution of an indirect pathway. These results suggest that, in quiescent animals, activation of the tension receptor enhances tibial flexion.

The present report brings to nine the number of multipolar receptor organs described and physiologically characterised for insect legs. These include the following: five individual neurones in the distal femur, which respond to joint position and movement (Coillot and Boistel, 1969; Kuster and French, 1983); an extensor tibiae muscle apodeme receptor sensitive to joint distortion (Bässler, 1977); the metathoracic accessory flexor TR, which is sensitive to accessory flexor muscle tension (present study); the pro- and mesothoracic flexor tibiae muscle tension receptors (Theophilidis and Burns, 1979); and the movement- and tension-sensitive coxo-trochanteral MRO (Bräunig, 1982; Bräunig and Hustert, 1985a,b). An additional nine multipolar sense organs have been described from cobalt fills of the tibial and tarsal nerves of the locust mesothoracic leg (Mücke, 1991), but nothing is known of their sensory physiology. The metathoracic accessory flexor TR neurone appears to be homologous to similar receptors in the prothoracic and mesothoracic legs, although its function may be more complex. Like them, it responds to the tension in a distal group of muscle fibres that act to flex the tibia (see Theophilidis and Burns, 1979). It differs in that (a) its primary dendrites generally do not branch, and (b) it is in a specialised muscle, the accessory flexor.

The ultrastructure and physiological characteristics of the accessory flexor muscle strongly suggest that it is of the slowly contracting, tonic type. The corresponding antagonist (M135c, accessory extensor) that attaches to the extensor muscle apodeme has essentially the same ultrastructure and thin:thick filament ratio (7:1, Hoyle, 1978). This muscle was shown by Hoyle (1978) to consist only of slow muscle fibres which display slowly rising contraction profiles with a slope proportional to stimulation frequency (e.g. 2.1 g per 10 s during 10 Hz stimulation of its single excitatory axon, SETi). Single twitches showed the same relaxation profiles as those observed here in single twitches of the accessory flexor muscle. The slow, tonic nature of the accessory flexor muscle must set constraints on its rate of tension development and hence on the range of afferent impulse frequencies developed in the tension receptor. The conclusion is that the system is operationally designed for the control of postural or slow to medium rates of limb movement, rather than fast phasic movements.

Although the commonest pattern of motor innervation recorded electrically from the accessory flexor muscle was three units (two slow excitors and one inhibitor), up to four motor axons were detected in this way, and up to five axons were seen in electron micrographs (Fig. 2C). It is probable that the full complement of efferent axons includes two slow flexor tibiae motor neurones (from the anterior and lateral groups), two common inhibitors (CI₂ and CI₃) and one dorsal unpaired median (DUM) neurone. The common inhibitory neurones in the locust are known to innervate slow muscle fibres exclusively. In tettigoniids, the accessory flexor muscle is innervated by 13 motor axons, but the presence of associated mechanoreceptors has not been reported (Theophilidis and Dimitriadis, 1990).

The metathoracic tension receptor, like those in the prothoracic and mesothoracic legs of the locust (Theophilidis and Burns, 1979), and in the legs of the stick insect (Bässler,1977), spikes tonically at rest. Passive stretch of the muscle into which the receptor inserts causes the receptor’s tonic firing rate to increase slightly. In the metathoracic leg of the locust, the relationship between passive accessory flexor muscle length and receptor firing rate is approximately linear, but in the prothoracic and mesothoracic legs it is not (Theophilidis and Burns, 1979). In the mesothoracic leg, it has been suggested that this non-linear relationship is caused by a corresponding non-linearity in the relationship between muscle length and tension (Theophilidis and Burns, 1979). The values of muscle length given by Theophilidis and Burns are for the entire flexor muscle, however, not for the fibres into which the receptor inserts. These fibres attach to the apodeme at an angle, so a linear movement of the apodeme (i.e. linear stretch of the entire flexor) will cause a non-linear change in the length of the ‘receptor’ fibres, thus probably contributing to the apparently non-linear length–tension relationship. In any case, the percentage change in TR spike frequency as the tibia is set at different angles in the range 0–160 ° is only about 6 % of its maximum spike frequency during isometric contractions of the accessory flexor muscle. We conclude that the TR is minimally important in coding static joint position. Golgi tendon organs of vertebrates also respond primarily to active contraction of their associated muscle fibres, and not to passive stretch (Jami, 1992).

Rapid stretches of the accessory flexor muscle generated transient increases in receptor firing rate (Fig. 3D,E). These presumably reflect transient transmission of tension to the proximal end of the muscle (where the receptor lies) that would rapidly be damped by the viscoelasticity of the fibres. Faster stretches caused larger transient changes in firing rate. Only very rapid movements over relatively large arcs caused substantial changes in firing frequency, so it seems unlikely that in natural situations the phasic response is particularly important (see Fig. 3Aiii,iv for more natural movements).

The metathoracic TR, like those of the anterior legs, responds much more strongly to active isometric tension than to imposed stretch of the muscle, indicating that tension, rather than length, is the primary stimulus. Artificial activation of a single excitatory flexor motor neurone is sufficient to affect the firing rate of the receptor. The relationship between the firing rate of the motor neurone and that of the receptor is linear, at least for the (lower) frequencies that could be tested. The accessory flexor/tension receptor system generally acts like a low-pass filter, smoothing out the more rapid changes in motor neurone firing frequency (see Fig. 5B). This is consistent with the conclusion reached above that the accessory flexor is primarily a slow, tonic muscle. Rarely, however, individual action potentials of an excitatory motor neurone caused clear changes in the TR firing rate (Fig. 5E), suggesting, perhaps, that the gain of the system can increase. Tension generated by the main part of the flexor muscle, or by the extensor muscle, is not registered by the TR. In other words, it responds only to tension generated by the accessory flexor fibres.

By combining our knowledge of motor neuronal connections (see references in Materials and methods) with the synaptic connections revealed in the present study, it is possible to propose a circuit diagram of the reflex effects of the TR (Fig. 10). Three salient points emerge. First, in quiescent animals, the TR gives primarily positive feedback onto the flexor tibiae motor neurones (Fig. 9), including those that innervate the accessory flexor. Second, negative feedback is also provided to the accessory flexor muscle (through the CIs), presumably to prevent instability of the positive feedback loop. Third, the extensor tibiae muscle receives reflex inhibition from the TR (through the actions of the receptor on SETi and CI₁), complementing the flexor excitation. The predominant effect of TR activity in quiescent animals is, as demonstrated by electrically induced contraction of the receptor muscle, to support tibial flexion.

In active animals (defined here as those attempting to struggle), the excitatory effects on flexor motor neurones were sometimes, but not always, replaced by inhibitory effects.

The common inhibitors exclusively innervate muscle units that receive slow motor innervation. They act to gate out this excitatory input and enhance the rate of muscle fibre relaxation. During rhythmic movements, the effect of this is to disable tonic fibres that might otherwise cause a build up of residual tension and thus increase joint stiffness. This permits fast motor units to dominate, and increases the velocity of the movement (Wolf, 1990). In the TR system, at least two mechanisms could operate under different conditions. First, and simplest, depending upon the relative synaptic efficacy of the various pathways, the common inhibitors could provide a negative feedback component to the positive feedback reflex loop represented by the excitatory connections between the TR and accessory flexor excitors. This could act to gate out the positive feedback. Second, the inhibitors could be driven centrally to relax both the accessory and main flexor muscles during rapid tibial extensions. Inhibition of the accessory flexor muscle during tibial extension should prevent it from building up tension and thus prevent an increase in TR firing rate. This mechanism may be particularly important during kicking and jumping (see below).

The accessory flexor/tension receptor system could act as a servomechanism if the slow motor neurones that innervate the main and accessory flexor muscles were driven centrally to produce a set posture. If the system were perturbed by passive tibial extension or an increased load, the TR would respond with increased firing. This would excite the flexor motor neurones and inhibit SETi. As long as the load or imposed extension remained, the positive feedback would continue (being limited by the maximal frequency of receptor firing). Once the muscle contraction overcame the load and the tibia re-flexed, then the receptor firing rate would decrease and the positive feedback subside. The responses illustrated in Fig. 9 demonstrate that such reflex circuitry is available to the locust, although we do not know whether it is used in this way. A similar proposal has been made for the presumably homologous receptors of the prothoracic and mesothoracic legs (Theophilidis and Burns, 1979).

During walking, the swing phase for the metathoracic leg is driven by tibial flexion (Burns, 1973). Because unopposed active tibial flexions do not cause marked increases of TR firing rate (not shown; see Theophilidis and Burns, 1979), the receptor must have only a minimal influence on normal swing. Obstruction of swing, however, would cause an increase of flexor tension and, thus, a pronounced increase in TR firing rate. A similar increase of receptor firing rate would occur if the flexor motor neurones began firing while SETi was still active at the end of stance or while the tarsus was still in contact with the substratum. The overall positive feedback reflex elicited by the tension receptor would enhance flexor firing and inhibit SETi, thus reinforcing the swing phase in the face of such perturbations.

The accessory flexor muscle, like the main tibial flexor muscle, should be strongly activated during the co-contraction phase of these motor programmes. Because the tibia is fully flexed, there should be a build up of isometric tension in the accessory flexor muscle and a large resulting increase in tension receptor firing rate. The accessory flexor tension receptor makes connections that are appropriate for it to contribute to continued flexor firing during co-contraction (i.e. it excites tibial flexor motor neurones) and to promote the final rapid extension (through excitation of the common inhibitors). The common inhibitors are depolarised throughout co-contraction, but particularly near the end (Burrows, 1995). It is not known what proportion of this depolarisation is derived from sensory inputs, although the rhythmic drive seen in CI₁ during walking is predominantly central in origin (Wolf, 1990). In any case, the role of sensory feedback in driving the jump and kick circuitry is still unclear (see Heitler and Bräunig,1988). Ablation of the accessory flexor muscle does not prevent jumping (A. P. Norman and T. Matheson, unpublished results) so, as for other sense organs, the animal must be able to compensate to some extent for the loss of any such feedback.

There are three clear questions that should guide further study of this system. First, what is the pattern of motor activity in the accessory flexor muscle during voluntary behaviours, particularly walking and the co-contraction phase of kicking and jumping? Second, what are the relative efficacies of excitor and inhibitor motor axons in regulating tension in the accessory flexor muscle? Third, how are TR reflexes gated centrally during voluntary behaviour?

How does this reflex organisation compare with that known for other leg multipolar sensilla? Are there common principles of operation unique to the type II sensilla, inasmuch as they are usually associated with muscles? Table 1 summarises the available information for stick insect and locust receptors. Two kinds of reflexes are seen in multipolar receptor systems. The first is a negative feedback reflex onto the main power muscle that stimulated the receptor through joint movement or tension (cxMRO, stick insect TR). This is usually accompanied by excitation of the antagonistic power muscle, and constitutes a resistance reflex. The second is a positive feedback reflex onto the homonymous or parallel power muscle (locust tibial flexor TRs, joint multipolar sensilla). There is generally inhibition of antagonistic power muscles, thus forming a load-compensating reflex circuit. Such circuits have also been found in muscle tension receptors of other arthropods (e.g. Macmillan, 1976). In addition, the positive feedback pathway of the locust metathoracic TR has a negative feedback component (Fig. 10). The negative component is mediated by the common inhibitors CI₂ and CI₃, as well as a long-latency inhibitory pathway onto the flexor motor neurones. It appears to be a specialisation of the locust metathoracic TR reflex circuitry, because it has not been found in other multipolar sensillum reflexes. Its function will remain unclear until we learn more about the conditions under which it becomes predominant. In principle, it could function as a high-threshold mechanism for gating out the positive feedback to the flexor muscle.

In summary, it appears that the multipolar receptors in insect legs have been adapted for a variety of reflex systems, some of which are devoted to feedback control of muscle tension in the face of variable load, while others are devoted to feedback maintenance of joint position (Bässler, 1977; Theophilidis and Burns, 1979; Bräunig and Hustert, 1985b; present study).

We thank Jan McKenzie for her help with the electron microscopy, and Jacques Hervé for his help with some of the initial experiments. Peter Bräunig, Malcolm Burrows, Fred Kuenzi and two anonymous referees made valuable comments on the manuscript. T.M. was supported by a research fellowship at Girton College, Cambridge, and by an SERC grant to Malcolm Burrows. L.H.F. was the recipient of an Erskine Fellowship from the University of Canterbury.