Adaptive changes in locomotor control after partial denervation of triceps surae muscles in the cat

Abstract

1. This report concerns a test of the hypothesis that gain in the stretch reflex pathway of cat medial gastrocnemius (MG) muscle during locomotion increases after denervation of its synergists, lateral gastrocnemius (LG), soleus (SOL) and plantaris (PL) muscles.

2. In four cats, MG, tibialis anterior (TA) and vastus lateralis (VL) muscles were implanted with electromyogram (EMG) electrodes. The cats walked on a row of elevated pegs, some of which were spring-loaded and could be triggered to pop up at the moment of foot touchdown, rapidly dorsiflexing the foot. Pre-stretch EMG activity in MG as well as short-, medium- and long-latency responses to the dorsiflexions were compared before and after unilateral denervation of synergists.

3. Short- and medium-latency responses of MG to perturbations increased in proportion to the increase in pre-stretch EMG in the days and weeks after partial denervation. This argues against an adaptive increase in stretch reflex gain independent of centrally generated extensor drive.

4. Local anaesthesia of the skin of the paw did not significantly change the sizes of the stretch responses of MG before or after partial denervation.

5. We conclude that pre-stretch EMG activity as well as stretch reflexes in MG muscle increased substantially after denervation of synergistic muscles. The data were consistent with an adaptive increase in central locomotor drive, causing more motoneuronal activity, which in turn resulted in an increase in the size of stretch reflexes.

Carrier et al. (1997) recently described adaptive changes in the control of hindlimb flexion in cats after denervation of ankle flexors. Certain aspects of these adaptive changes persisted after the cats were spinalized, suggesting that enduring changes in segmental transmission had occurred. Since the adaptive changes did not develop in cats spinalized before the partial denervation, Carrier et al. (1997) suggested that signals descending from supraspinal centres were responsible for inducing adaptive changes in the wiring of spinal interneuronal networks associated with the locomotor pattern generator. As the workings of the pattern generator are strongly influenced by sensory inputs (Pearson, 1995; Rossignol, 1996), one might suspect that changes in sensory pathways could be involved in these adaptations. Indeed it has been shown that operant conditioning in rats can be used to produce long-term augmentation or attenuation of transmission in stretch reflex pathways and that descending input mediated by the corticospinal tract is necessary for the relevant changes to occur in the wiring of the spinal cord (Chen & Wolpaw, 1997). There are numerous mechanisms by which sensorimotor transmission may be modulated in both the short and long term, including fusimotor action (Prochazka, 1996), presynaptic inhibition (Rudomin, 1999) and the action of neuromodulators such as serotonin (Rossignol et al. 1998). Finally, there is the practical question of whether intensive training could be used to augment desirable adaptive changes in spinal motor responses in spinal cord-injured people (Carrier et al. 1997).

In experiments that followed on from those of Carrier et al. (1997), Pearson et al. (1999) reported large increases in stance-phase MG EMG activity following denervation of its synergists, LG, SOL and PL. Early (pre-ground-contact) and late (mid-stance) components of EMG both increased after denervation, the latter within a day and the former more gradually over several days. Because the late EMG components were correlated in size with the amount of muscle stretch, Pearson et al. (1999) suggested that they were mediated by afferent signals generated soon after the onset of stance. It was suggested that partial denervation led to an adaptive facilitation of transmission of these afferent signals in reflex pathways, reinforcing the central drive to MG motoneurons. The authors argued that their previous results ruled out changes in the monosynaptic group Ia pathway, but they hinted that group Ib tendon organ afferents might be involved. Considerable importance was attached to the slower time course of the increase in the early (pre-contact) component of EMG activity. It was suggested that changes in the early components reflected a gradual re-scaling of internal models of muscle stiffness that in turn depended on error signals provided by group Ia afferents.

Our study was designed to test some of these ideas. In normal cats, before denervation, rapid ankle dorsiflexions were applied at the onset of the stance phase of step cycles to elicit EMG stretch responses in MG. The day after denervation of LG, SOL and PL, these responses had increased in size, as had the centrally generated pre-ground-contact EMG. Over the ensuing days and weeks pre-contact and reflex responses increased further. There was no significant difference in the time course of these respective increases. The results are consistent with adaptive augmentations of extensor locomotor drive that in turn cause increases in the size, but not the gain, of stretch reflexes.

METHODS

The experiments were performed on four cats with the ethical approval of the University of Alberta Health Sciences Animal Welfare Committee. They conformed to the guidelines of the Canadian Council on Animal Care.

Surgical implantation

In a single 2-3 h aseptic procedure under pentobarbitone anaesthesia (35 mg kg⁻¹ intraperitoneal, maintained intravenously), EMG wires (632ss Cooner, Chatsworth, CA, USA) were sewn into MG, TA and VL muscles of each hindlimb. The intramuscular portions of the wires were deinsulated over about 4 mm. The wires were led subcutaneously to the cat's head where they were embedded in an acrylic headpiece secured to the skull by three bone screws. Incisions were closed with 4/0 nylon monofilament sutures. Buprenorphine (0.03 mg kg⁻¹) was administered subcutaneously: this powerful analgesic has an effect lasting about 12 h. The animals were placed on blankets in a heated enclosure. Typically they were conscious and moving around within 6-10 h and showed no obvious sign of pain or discomfort. A second similar dose of buprenorphine was given 10 h after the first to guarantee a comfortable and pain-free recovery.

Denervation

Denervation of the muscles was carried out following the methods of Opearson et al. (1999). In a 45 min aseptic procedure conducted under halothane anaesthesia, an incision was made proximal and dorsal to the knee joint on the medial side of the left hindlimb. Nerves to LG, SOL and PL were identified with the help of electrical stimulation and cut. The incision was sutured shut and the cat recovered with buprenorphine analgesia.

Locomotor trials

Cats walked along a 2 m-long walkway for food rewards (Fig. 1). The walkway comprised 12 vertical pegs, 18 cm in height, each offering a 3 cm-diameter flat surface for foot support. Four of the pegs were spring-loaded with an electromagnetic release mechanism. Three photoelectric sensors mounted less than 1 mm above the support surfaces of each active peg detected the moment the cat's paw interrupted light directed across the surface from corresponding photodiodes on the opposite side. This triggered a solenoid, releasing the spring-loaded portion of the peg, which moved upwards by 1-1.5 cm, rapidly dorsiflexing the cat's foot. Figure 1B shows stick figures obtained from 30 frames s⁻¹ video films of a cat 1 day before and 1 and 12 days after partial denervation, walking on a locked peg (top panels, no stretch) and a peg that popped up at the moment of foot contact (bottom panels). Springs of three grades of stiffness were available to produce peg velocities of 250, 450 and 1000 mm s⁻¹ corresponding to triceps surae stretch velocities of 50, 90 and 200 mm s⁻¹, respectively, assuming a lever ratio around the ankle joint of 1:5. Each active peg was equipped with custom-made length gauges to monitor peg displacement. Force transducers were embedded in the support surfaces of the active pegs to measure reaction forces but unfortunately the signals depended crucially on where the paw was placed on the surface, so these measurements were not considered in this report. EMG responses were recorded in MG, TA and VL.

Figure 1. Setup diagram and stick figures of cat hindlimb.

A, schematic diagram of the walkway. Any one of 4 spring-loaded pegs could be triggered to release at the moment of foot contact, which was detected by light sensors mounted on the top of the loaded peg. B, stick figures constructed by tracing hip, ankle and toe markers from video still frames (the toe markers were on the lateral side of the foot, so they are slightly above the contact surface). Knee joint position was inferred by assuming constant hip-knee and knee-ankle segment lengths. Top row, unperturbed trials; bottom row, peg-popping trials. Each column represents data collected on a different day. Left, 1 day before denervation of lateral gastrocnemius (LG), soleus (SOL) and plantaris (PL); middle, 1 day after denervation; right, 12 days after denervation. For each set the 4 stick figures were taken from video frames corresponding from left to right to 30 ms before foot touchdown, touchdown, 30 ms after touchdown and 120 ms after touchdown.

Foot pad anaesthesia

To determine the effect of cutaneous input on EMG responses, two animals were anaesthetized with halothane, and 1 ml of 1 % lidocaine hydrochloride (lignocaine) was injected into the paw pads of the partially denervated hindlimb. The animals recovered from the anaesthesia within about 20 min and data collection resumed. Foot anaesthesia was monitored continuously and data collection was discontinued at the first sign of reactions to pin prick in the locally anaesthetized foot.

Terminal surgery

In order to verify the denervation at the end of the experiment, cats were anaesthetized with pentobarbitone (40 mg kg⁻¹ intraperitoneal). Nerve branches innervating triceps surae and the stumps of the severed nerves were dissected, electrically stimulated and any resulting contractions were identified visually. The animals were then killed with an overdose of pentobarbitone. Triceps surae and PL muscles were removed from both limbs and weighed.

Data collection and analysis

EMG signals were amplified (×500-2000), high-pass filtered (30 Hz), full-wave rectified and low-pass filtered (1000 Hz). EMG and peg displacement signals were digitized (2000 samples s⁻¹) and stored using Signal 1.82 software, a CED1401 interface (Cambridge Electronic Design, Cambridge, UK) and a personal computer. The data were analysed off-line using Matlab 5.0 software (Mathworks Inc., Natick, MA, USA). Data from individual steps were aligned to the photoelectric contact signal, whose time of occurrence we defined as t₀. This signal triggered the solenoid to release the peg. The first detectable peg displacement commenced at about (t₀+ 5) ms. We estimated that the earliest possible latency of monosynaptic reflex responses was therefore (t₀+ 10) ms (Prochazka et al. 1976). Accordingly, MG EMG responses were divided into four bins, representing pre-stretch EMG, and short-, medium- and long-latency reflexes (corresponding to M1, M2 and M3 stretch reflexes in monkeys (Lee & Tatton, 1975) and cats (Ghez & Shinoda, 1978)). The pre-stretch bin (‘a’ in Fig. 2) encompassed the pre-contact build-up of MG EMG and terminated at (t₀+ 10) ms. The onset time of pre-contact build-up was computed in each individual trial as follows. First, the background mean (m_b) and standard deviation (σ_b) of MG EMG were computed in mid-swing over a 50 ms interval centred at (t₀ - 200) ms. The moving window mean (m_w) and standard deviation (σ_w) of the EMG signal were then computed over consecutive 10 ms intervals starting centred at (t₀ - 40) ms and moving back in time in 5 ms increments. When m_w and σ_w fell below the criterion levels defined below, the time in the middle of the window was taken as the onset time: (1) 100(m_w - m_b)/m_w < 20 (difference between moving window mean and background mean < 20 %); (2) 100(σ_w - σ_b)/σ_w < 40 (difference between moving window s.d. and background s.d. < 40 %).

Figure 2. Examples of averaged EMG of MG from 1 animal before and after the denervation.

A, normal steps on pegs recorded 1 day before (left) and 12 h after (right) the denervation surgery. B, steps 1 day before (left) and 12 h after (right) partial denervation with a medium-speed stretch applied at the moment of foot touchdown. Bottom trace, mean EMG; middle trace, peg displacement signal; top trace, change in firing rate (impulses s⁻¹) of Ia afferents modelled from the displacement signal. Dashed lines demarcate time intervals a, b, c and d, corresponding to pre-stretch, short-, medium- and long-latency components of MG EMG. Sweeps were aligned to the foot-contact signal at time zero (t₀).The mean onset time of bin a shown by the dashed lines is the mean of individual onset times computed in each trial according to specific criteria (see text). Data are from cat 3; each panel represents > 12 trials.

The short latency bin (‘b’) spanned the 20 ms following the termination of the pre-stretch EMG (i.e. +10 to +30 ms), the medium latency bin (‘c’) the next 35 ms (+30 to +65 ms) and the long-latency bin (‘d’) the final 100 ms (+65 to +165 ms). The long-latency bin was chosen to correspond approximately to the late component of Pearson et al. (1999; note that these authors did not use a fixed latency from foot contact, but rather centred the long-latency component around the peak EMG in the stance phase).

The mean EMG within each bin was computed for an average of 15 steps in a given experimental condition. EMG varied in absolute amplitude from one cat to another due to electrode characteristics, so to give the data from each cat equal weight, all EMG signals from a given cat were normalized to the peak of the averaged step cycle EMG profile obtained from 45 unperturbed steps prior to denervation. A set of stretch trials was performed in each cat during quiet stance. In these cases peg displacement was triggered manually. Because there was steady tonic EMG activity prior to stretch in these static trials, the pre-stretch bin, a, was set to encompass the interval -95 to +10 ms with respect to the trigger signal, corresponding to the mean span of bin a in the locomotor trials. Bins b and c spanned the same intervals as in the locomotor trials.

RESULTS

Figure 2 shows mean stance phase MG EMG profiles recorded from one animal for normal (A) and perturbed (B) walking before (left) and 12 h after (right) denervation. In Fig. 2B, peg displacement signals are shown; in these cases medium-stiffness springs were used, which produced 450 mm s⁻¹ peak peg velocity, corresponding to 90 mm s⁻¹ peak stretch velocity of triceps surae muscles. It is assumed here that in the 25 ms it took for the peg to pop, little rotation of the tibia occurred and therefore most of the motion of the paw was transferred by lever action to the triceps surae muscles acting about the ankle. A frame-by-frame analysis of video films supported these assumptions (Fig. 1B).

As the springs were not quite stiff enough to produce maintained step displacements or perfectly matched displacement signals before and after denervation, we felt it important to estimate the responses of muscle spindle Ia afferents to the actual displacements. Profiles of modulation of Ia ensemble firing rate were computed from the displacement signals using Ia models selected from the literature and implemented in Matlab Simulink software (Prochazka & Gorassini, 1998). The upper traces in Fig. 2B show profiles generated by the following model: Ia firing rate modulation = 4.3 × velocity^0.6+ 2 × displacement. The profiles were surprisingly transient in nature, the relatively slow decline in length after the initial jump being enough to return the Ia rate close to background levels. From the point of view of our experiment, the estimated Ia profiles showed only minor differences before and after denervation. The fact that they were transient and not maintained was important in relation to the interpretation of long-latency responses. Similar profiles were obtained with the Hasan model (Prochazka & Gorassini, 1998).

In Fig. 2 it is clear that pre-stretch EMG and short-, medium- and long-latency EMG components all increased after denervation in both unperturbed (Fig. 2A) and perturbed (Fig. 2B) steps.

Figure. 3 shows mean EMG responses of MG, VL and TA to 200 mm s⁻¹ stretches before (A) and after (B and C) denervation in three cats. The number of trials in which EMG recordings were available from all three muscles in each of the three cats varied from 5 to 17, so to maintain equal weighting, each EMG value in Fig. 3 was computed from the mean of three individual EMG means. Each individual mean EMG was normalized to the muscle's peak mean EMG in unperturbed step cycles on three consecutive days prior to partial denervation.

Figure 3. Mean stretch responses in MG, VL and TA before and after partial denervation.

Mean EMG responses of MG, vastus lateralis (VL) and tibialis anterior (TA) to 200 mm s⁻¹ stretches before partial denervation (A), 1 day after partial denervation (B) and 6 days after partial denervation (C). The transient responses in all three muscles were much more prominent than in Fig. 2 because the stretch rates were more than twice as large. Note the similar latencies of MG and VL responses and the longer latency of TA responses. MG and VL responses had increased, while TA responses had declined, 6 days after denervation. Data are from cats 1, 2 and 4. Normalization and averaging are described in the text; n = total number of trials per condition.

The speed of the stretches was more than double that in Fig. 2B, and the transient responses in all three muscles were therefore very prominent, which explains why the pre-stretch EMG levels appear so small compared to those in Fig. 2. Interestingly, there were large short-latency responses in VL as well as MG and delayed responses in TA that were essentially reciprocal to those in MG. The mean MG and VL responses had increased 6 days after denervation whereas that in TA had significantly decreased (see Discussion).

Figure. 4 shows the time course of the change of short-latency (A), medium-latency (B) and long-latency (C) responses to perturbations before and after denervation in four cats. Again, to ensure equal weighting per cat, each point in Fig. 4 represents the mean of individual means (±s.e.m.) of 10-22 normalized EMG responses obtained in each of the four cats. Each panel also shows the time course of the change of the mean amplitude of pre-stretch EMG. The first column shows data from steps on locked pegs (unperturbed steps). In this case triceps surae stretch was associated with the yield during weight bearing at the onset of the stance phase of the normal step cycle. The second, third and fourth columns show data from trials in which pegs popped, stretching triceps surae at estimated speeds of 50, 90 and 200 mm s⁻¹, respectively. The first three points in each plot in this figure correspond to the mean normalized MG EMG within the relevant time bin on each of 3 days before denervation of synergists. Comparing these three values across the four top panels, the short-latency responses increased with stretch velocity, as would be expected if they were mediated by group I muscle afferents.

Figure 4. Mean normalized MG EMG data from all animals recorded over a 22 day period after the denervation.

Each column displays data from trials with different speeds of muscle stretch. Left to right: unperturbed steps (unknown stretch rate), and 50, 90 and 200 mm s⁻¹ stretch. A-C, plots of the time courses of short-, medium- and long-latency bins defined in Fig. 2 (•) and corresponding pre-stretch bins in the same trials (○). The first three pairs of data points in each graph, to the left of the vertical dashed lines, represent data collected on the 3 days before denervation. Error bars, ±1 s.e.m. Note the similar time course of pre-stretch and stretch-response components in all graphs. Each data point represents 64 trials, 4 cats.

Mean pre-stretch EMG amplitude averaged over all trials had risen by at least 60 % 2 days post-denervation and by about 140 % 10 days post-denervation. There was no significant difference between pre-stretch EMG in trials with and without perturbations, indicating that the cats were not anticipating which peg would pop. The short, medium- and long-latency EMG response components all increased in parallel with the mean level of the pre-stretch components. This was true of the large increases on the day after denervation (day 1) as well as of the more gradual increases on subsequent days (note that Pearson et al. 1999 restricted their comparisons to day 1 onwards). The duration of pre-stretch components of MG EMG also increased abruptly after denervation (Fig. 5) but thereafter it slowly declined.

Figure 5. Mean duration of pre-stretch components of MG EMG before and after the denervation.

The onset and termination of pre-stretch components were identified quantitatively and included a 10 ms period after foot contact (see Methods). Note the immediate increase in duration from 90 ms to around 120 ms the day after denervation and a slow decline thereafter. Each bar in the histogram represents 256 trials (means ±s.e.m.), 4 cats.

As the mean amplitudes of all EMG components, including the pre-stretch EMG, increased more or less in parallel after denervation, we were interested to know whether the ratios of the stretch reflex components to pre-stretch components (i.e. the reflex gain) increased after partial denervation. Using the set of single-sweep data that gave rise to Fig. 4, we computed all the gain ratios per condition per cat and their means ±s.e.m. The means of these sets of means (1 per cat per condition) were computed and are plotted in Fig. 6. With the exception of the 200 mm s⁻¹ short- and medium-latency data, there were no consistent trends in the ratios over time after denervation, i.e. the stretch responses retained the same proportional relationship to pre-stretch components in the 3 days before denervation of synergists as well as in the days and weeks after denervation. The only significant trends were seen in the short- and medium-latency responses to 200 mm s⁻¹ stretches. In these cases, there was a 25-30 % fall in the ratios immediately after denervation, which was maintained for the 3 week duration of the study (this is attributable to the reduced afferent input from synergists: see Discussion). If there had been a sudden increase in the reflex responses with a more gradual increase in the pre-stretch component (Pearson et al. 1999), this would have shown up as an increase in the ratio on days 1-4 followed by a gradual decline. In other words, there is no compelling reason to think that there had been a sudden adaptive increase in the gain of spinal stretch reflexes in the first few days after denervation. Rather, the increases in the amplitudes of the stretch responses can be explained by simple scaling of reflex responses to the increased background (pre-stretch) EMG after denervation.

Figure 6. Time course of reflex gain: means of ratios of individual stretch reflex to pre-contact components.

Same set of data as in Fig. 4 In most cases, the mean ratios remained the same or decreased after denervation (i.e. pre-stretch and reflex components increased with a similar time course after denervation and reflex gain did not increase significantly). Means ±s.e.m. from all 4 cats.

We performed a similar analysis in three cats of stretch reflexes elicited by triggering peg displacement as the cats stood still on them. The difference in these trials was that by slightly adjusting the posture of the cat on the peg by varying the position of a food reward, background EMG levels could be matched reasonably well across trials. The left panels in Fig. 7 show that under these conditions pre-stretch EMGs as well as short- and medium-latency MG EMG responses to 200 mm s⁻¹ stretches did not change significantly after partial denervation. Reflex gain, estimated as the mean ratio of stretch EMG responses to pre-stretch EMG (Fig. 7, right panels) showed a 25-30 % fall in short-latency components as it had in corresponding stepping trials (Fig. 6A, right) and stayed fairly constant for medium latency components (Fig. 7, right panels).

Figure 7. Responses to rapid (200 mm s⁻¹) perturbations applied during static stance.

Left panels, time courses of short- (A) and medium-latency (B) bins (•) and corresponding pre-stretch bins in the same trials (○). The first three pairs of data points in each graph, to the left of the vertical dashed lines, represent data collected on the 3 days before denervation. Error bars: ±1 s.e.m. Right panels, mean ratios of reflex to pre-contact components; same set of data as in the left panels. The mean ratios did not show a trend to significantly increase from day 1 after denervation. Data from cats 1-3, 6-17 trials per cat.

Local anaesthesia of the footpads was performed in two cats. In cat 1 it was done 33 days after denervation. In cat 2 it was done 2 days prior to denervation as well as 61 days after denervation. Complete insensitivity was tested by applying pinprick stimuli to several locations on each pad. After a few minutes of complete footpad anaesthesia the cats displayed no obvious abnormalities in locomotion overground or on the pegs. Figure 8 shows mean EMG (A, TA; B, MG) for steps before and during footpad anaesthesia for stretches of 90 and 200 mm s⁻¹. The non-anaesthesia control trials (thin lines) were done the day before the anaesthesia trials (thick dashed lines). There was remarkably little difference between matched trials, suggesting that the results were valid (different cats) and reliable (different days).

Figure 8. Stretch responses in TA and MG before and after partial denervation.

Mean EMG responses of TA (A) and MG (B) in cats 1 and 2 to 90 and 200 mm s⁻¹ perturbations before (control; continuous thin lines) and during (thick dashed lines) footpad anaesthesia. The control trials were done the day before the anaesthesia trials. There is remarkably little difference between corresponding trials with and without local anaesthesia.

Figure 9 summarizes the detailed statistical comparisons: the mean ±s.e.m. of pre-stretch, and short-, medium- and long-latency EMG components are shown before (□) and during footpad anaesthesia (▪). There were small, but statistically significant (P < 0.05), differences in two pairs of pre-stretch components before denervation in cat 2 (Fig. 9B and C, left). Statistical significance was also reached in the long-latency responses (both cats, 200 mm s⁻¹; Fig. 9C, right). Overall, however, the data suggest that skin input from the footpads did not contribute significantly to the EMG responses to perturbations. This does not exclude skin input from other parts of the foot or leg playing an important role (see Discussion).

Figure 9. Normalized mean MG EMG data from normal and foot-anaesthetized walking conditions.

Left column, data from cat 1 before partial denervation; right column, combined data from cat 1, 32 days after denervation and cat 2, 61 days after denervation. A, unperturbed steps; B, 90 mm s⁻¹ muscle stretch; C, 200 mm s⁻¹ muscle stretch. In each graph means ±s.e.m. of four EMG components are plotted for normal (□) and anaesthetized (▪) conditions. Two pairs of values in cat 2 (pre-stretch, pre-denervation) showed small differences that just reached significance (B and C, left; *P < 0.05). One pair of post-denervation values (both cats) showed a large and significant difference (C, right; *P < 0.01).

DISCUSSION

As shown by Pearson et al. (1999), pre- and post-contact EMG activity in MG during unperturbed stepping increased following denervation of the synergistic muscles LG, SOL and PL. Our experiments were designed to evaluate and compare the time course of the centrally generated (pre-contact) and reflex-mediated (post-contact) components of this adaptation. Short- and medium-latency components of the response to perturbation increased with increasing speed of the applied displacement of the foot, which suggests that stretch reflexes contributed to them. Increases in these components in the days following denervation (Fig. 4) therefore support the hypothesis that stretch reflexes increase in absolute amplitude after partial denervation. The timing of the short latency responses and the lack of effect of local anaesthesia of the paw on them suggest that they were mediated by group I muscle afferents.

The long-latency components of MG EMG did not change appreciably with a fourfold increase in stretch speed (see Fig. 4C, compare first three data points in each panel: no significant increase as stretch speed increased from 50 to 200 mm s⁻¹). This was surprising, not least because the reflex gain hypothesis we were testing was derived from late components of EMG in the step cycle rather than early components (Pearson et al. 1999). However, it became clear from the brief time course of the modelled Ia responses to the stretches (Fig. 2B) that Ia input was unlikely to contribute significantly to the long-latency EMG responses in MG in our experiments. Given that the MG muscle was active and remained stretched by the perturbations throughout the long-latency EMG responses, tendon organ Ib firing was probably elevated during this time (Prochazka & Wand, 1980) and may have contributed to these long-latency components. If Ib reflex gain was significant, the only way we can explain the lack of correlation between the speed of peg motion and the size of the long-latency components is that within 160 ms of stretch onset, ground reaction force and therefore MG Ib firing reached the same steady-state value determined by the gravitational load of the cat's body weight rather than the speed of stretch.

The main new finding in our study is that short- and medium-latency stretch reflex responses to the perturbations increased in proportion to the increases in pre-stretch EMG. There was little evidence in our data for the notion that MG reflexes increased sooner after denervation than the centrally generated pre-stretch EMG components. The proportional scaling of the amplitude of stretch reflexes to ensemble motoneuronal activity is a well-known property of reflex transmission (Marsden et al. 1976). Actual increases in the gain would require increases in the ratios of reflex to pre-stretch components (i.e. increases in the constants of proportionality). These were not seen after denervation (Fig. 6 and 7). The long-latency EMG components also increased after denervation in proportion to pre-stretch EMG, but as just mentioned, the peripheral contribution to these components is hard to estimate. However, if Ib reflexes were involved, these were scaled to the pre-stretch activity too, so again there was no reason to invoke reflex gain changes after denervation.

Some important differences between our experiments and those of Pearson et al. (1999) should be noted. First, our cats were walking on pegs rather than on a treadmill. It has been suggested that peg-walking is more demanding and that this might have led to a different mode of locomotor control. We doubt this, because the cats negotiated the pegs with ease after just one or two trials and ran back and forth across them for food rewards. Second, the largest adaptive increases in EMG components occurred on the first day after denervation, compared to control trials on the 3 days prior to denervation. However, there is a problem in comparing these values directly. Denervation of LG, SOL and PL abolishes afferent input from these muscles and therefore some of the heteronymous synaptic input to MG motoneurons. About 20 % of total Ia monosynaptic input to MG can be assumed to be absent after the denervation (Burke & Rymer, 1976). This is why Pearson et al. (1999) only considered adaptive changes from baseline measurements commencing a few hours after denervation. However, even if we exclude our pre-denervation data, our conclusions regarding parallel and proportional increases in pre-stretch and reflex components after denervation still hold. In fact in Fig. 6A, right panel, the ratio of short-latency responses to pre-stretch activity dropped immediately after denervation in 200 mm s⁻¹ stretches, consistent with the estimated 20 % drop in Ia synaptic input. Third, the hypothesis of Pearson et al. (1999) regarding a rapid increase in reflex gain followed by a more gradual re-scaling of central locomotor drive was based on changes in late components of EMG, by which the authors meant components of EMG occurring within a 100 ms period centred on the peak of EMG activity in the step cycle. This is an important distinction: as we were specifically testing the reflex gain hypothesis, we defined long-latency components as those occurring within the interval 65-165 ms after the abrupt perturbations coinciding with foot contact. In contrast, the late component of Pearson et al. (1999) had no fixed latency with respect to foot contact, as the peak of MG EMG occurred at different times in the step cycle. Thus our conclusions regarding responses to known test inputs do not contradict the observations of Pearson et al. (1999) though they are at variance with the hypothesis drawn from them.

Carrier et al. (1997) compared patterns of treadmill locomotion of intact and spinal cats before and after denervation of ankle flexors. Cats with intact spinal cords adapted to denervation within a few days and achieved symmetrical gait by increasing hip and knee flexion. After spinalization knee hyperflexion persisted and gait was asymmetrical. Yet a cat first spinalized and then partially denervated regained symmetrical gait. The results suggested that there were gradual changes in both supraspinal and intraspinal connections after partial denervation. Because pre-contact EMG in MG is largely centrally generated (Engberg & Lundberg, 1969; Gorassini et al. 1994), adaptive increases in descending drive after denervation of synergists was probably responsible for the immediate increases in pre-contact EMG in our cats. We did not find convincing evidence for a change in reflex gain independent of the changes in pre-stretch activity, i.e. there was no evidence of a reorganization of spinal reflex transmission in MG after denervation of its synergists.

Cutaneous receptors in and around the footpads signal ground contact with high-frequency bursts of discharge (Trend, 1987). In perturbation trials, the sudden upward force of the pegs on the footpads presumably caused very rapid bursts of firing in many of these receptors. Synchronous activation of cutaneous nerves by electrical stimulation during locomotion evokes short-latency responses in ankle muscles in normal cats (Abraham et al. 1985). Mechanical stimuli applied to the dorsum of the foot can also evoke EMG responses, particularly during the swing phase (Wand et al. 1980). Short- and long-latency cutaneous excitatory and inhibitory reflexes have also been recorded during fictive locomotion in cats (Schmidt et al. 1989; LaBella et al. 1992; Degtyarenko et al. 1996). In humans it has been argued that medium-latency responses are mediated purely by skin input (Corden et al. 2000). Yet it has also been shown in intact cats that the abolition of skin input does not seem to have a major effect on locomotor control (Engberg, 1964; Bouyer & Rossignol, 1998). In two cats we used local anaesthesia to abolish input from footpad skin receptors (Fig. 8 and 9). This had no detectable effect on any of the components of the EMG response to peg pops. It is true that other skin receptors of the foot and lower leg probably continued to respond to the perturbations and it is quite possible that these receptors contributed to the short- and medium-latency EMG responses before and during footpad anaesthesia. On balance, however, the evidence favours muscle receptor input as being the dominant influence. Note that our test of the hypothesis of adaptive changes in stretch reflex gain does not depend on the type of receptor involved, nor indeed did the original paper insist on the responses being mediated by muscle receptors (Pearson et al. 1999).

Could any of the components of the response have been auditory startle reactions to the peg pops? To address this, we used a sensitive Eletret microphone placed in the walkway at a position corresponding to the cat's head to record the sound caused by one of the pegs popping at its fastest velocity. This showed that though there was very little sound at the onset of peg movement, a sharp sound was generated when the peg reached its stop about 25 ms after release. This second sound was of a similar intensity to a sharp handclap about 30 cm from the microphone. This could have resulted in startle responses, at least in the first few trials. The latency of auditory startle responses in cat hindlimbs is 18-20 ms (Gruner, 1989) giving an expected net latency of startle responses of 43-45 ms in our trials, which would correspond to component c. As mentioned above, the long-latency component d was unaltered in trials with peg popping compared to non-peg pop trials, so neither startle nor stretch reflexes need to be invoked for these components. Components a and b precede the sound pulse so a startle response contribution can be ruled out for these components too. Auditory startle responses generally show rapid adaptation whereas response component c showed little adaptation after many repetitions in our experiments. Therefore, although we cannot rule out a contribution of auditory startle responses to component c in the fastest perturbations, proprioceptive inputs seem the more likely source.

Could displacements about joints other than the ankle have contributed? The video films and the stick figures in Fig. 1 show that in the first 30 ms the displacement of the peg caused dorsiflexion about the ankle but very little change in the other joints. The final frame shows that on day 1 the leg yielded and by 120 ms the hip joint had been displaced. In principle this might have affected the long-latency component (d) but in fact this component remained constant whether there were perturbations or not. Admittedly the large responses in VL in Fig. 3 do suggest that the knee extensors were transiently stretched, though the responses could also have been mediated by ankle extensor afferents with synaptic connections to knee extensor motoneurons. We did not attempt a full analysis of the changes in VL after the partial denervation because these were not of direct relevance to the hypothesis being tested. The large delayed responses in TA, which were reciprocally timed with peaks of activity in MG (Fig. 3), might have been caused by a small transient re-stretch of TA at the end of the step perturbation. It is interesting that these responses became much smaller in the days after partial denervation. Because we cannot be certain of the sensory origin of the TA responses and because their longer latency allowed time for more complex processing, we have not attempted to explain this decline. It therefore remains an open question whether descending inputs and/or spinal reflex transmission to TA motoneurons change in the days after partial denervation of triceps surae muscles.

We conclude that the centrally generated pre-contact MG EMG activity increases in an adaptive manner after denervation of synergistic muscles and this is associated with an increased amplitude of stretch reflexes. Both mechanisms serve to increase the force produced by medial gastrocnemius to counteract the loss of force from the denervated muscles. There was no compelling evidence of a change in the gain of reflex transmission after denervation.