| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1Department of Neuroscience, University of Minnesota, Minneapolis, Minnesota 55455; and 2Department of Neuroscience, Human Physiology Section, University of Rome at Tor Vergata and Scientific Institute Santa Lucia, 00179 Rome, Italy
Submitted 4 March 2003; accepted in final form 20 July 2003
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
). Their activity is linearly related to the limb axis length and orientation for passive postures and movements of the hind limb (Bosco et al. 1996; Poppele et al. 2002). This implies that the activity encodes the limb endpoint kinematics rather than local kinematic information about limb geometry or specific joint angles. We tested this experimentally for passive limb postures by mechanically decoupling limb endpoint position from limb geometry by means of joint constraints (Bosco et al. 2000). The result was that the activity of only about half of the cells tested was unaltered by the constraint. The activity of the other half also related to the limb axis parameters in the constrained condition, but the relationship was different from that seen for the unconstrained condition.
Recent investigations of DSCT behavior during continuous limb movements provided evidence that the changes evoked by joint constraints were due to a modification of specific length and orientation response components (Bosco and Poppele 2002). These components are expressed separately and independently in the DSCT activity during passive limb movements (Poppele et al. 2002), and the same response components accounted for the responses to both the constrained and unconstrained movements. This finding implies that the changed responses resulted from a re-weighting of response components representing limb axis length and orientation. The re-weighting would lead to an altered representation of limb axis length and orientation without introducing new or different response components representing local limb geometry explicitly.
This finding did not, however, rule out a major role by the limb biomechanics in determining this behavior. The question remains about how such modifications might occur. In the accompanying paper (Bosco et al. 2003), we showed that neuromodulators like serotonin can alter the relative strengths of the specific response components related to limb axis length and orientation. Thus one possibility is that the sensory input may itself modify synaptic weightings in a similar way. Alternatively, the modifications we observed with joint constraints may have a biomechanical basis resulting from specific differences in the constrained and unconstrained limb.
The current study was undertaken to examine this issue by comparing the biomechanical consequences of two types of limb perturbations on the resulting changes in DSCT activity. The perturbations we used to modify the sensory input were either joint constraints or muscle stimulation applied while the limb was moved along a given foot path. In this way, the limb axis length and orientation were unchanged while the associated limb geometry and/or joint torques were significantly altered. Presumably this also altered the resulting sensory input from the limb. We found that the kinematic alterations in the joint angle trajectories and the changes in joint angle covariance did not correlate with the observed changes in the DSCT responses to the movement.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
) and Poppele et al. (2002). A computer-controlled robot arm (Microbot AlphaII+, Questech, Farmington Hills, MI) moved the ipsilateral hindlimb limb through a footpath modeled on a step cycle (see Fig. 1A). The kinematic profile of the foot trajectory was comparable to that produced in normal cat walking (see also Bosco and Poppele 1999). The step cycle duration was 2.83 s or about half the cycle duration reported in Poppele et al. (2002)
|
We recorded unit activity from DSCT neurons during the step-like movements in two basic conditions, a "control" condition where the limb geometry was determined by the passive limb mechanics and a "test" condition where the passive limb geometry was perturbed either by joint constraints or by muscle stimulation.
Joint constraints.
We applied two types of joint constraint designed to dissociate limb geometry from foot kinematics. One was a rigid constraint, namely a Plexiglas strip mounted between surgically implanted pins in the femur (
5 cm from the femur head) and tibia (6 cm from its distal end). This constraint limited or prevented angular motion of the knee joint. The other was an elastic constraint (represented by the wavy line in Fig. 1A) applied between the tibia pin and the robot platform, (see Bosco et al. 2000 for more details). This constraint mostly limited the angular motion of the shank elevation. We refer to the former as a knee constraint and the latter as an ankle constraint.
Muscle stimulation.
We activated hindlimb muscle groups by stimulating dissected ventral roots (VRS) as we described previously (Bosco and Poppele, 1997). In brief, we isolated two VR filaments that activated separate muscle groups in the anterior or posterior hindlimb, respectively, in each animal. Usually stimulating an individual rootlet activated primarily one muscle and to a lesser extent, other, mostly functionally agonist muscles. Although we made no attempt to quantify the actual contractions, we did observe the contractions and verify that each cell responded to the contractions.
We used two stimulation paradigms that each employed 0.1- to 0.5-ms shocks that were above threshold for a visible muscle contraction. One stimulus was a pseudorandom activation at a mean rate of 8/s. The random stimulus intervals were effective in producing a stable response without fatigue for a duration of 
1 min. This stimulus was applied while the limb was not moving and the limb axis was in a vertical orientation. It was used to determine the effect of the muscle twitches on DSCT activity (Osborn and Poppele, 1983) and to provide an independent control for stimulus effectiveness for each recording. The other paradigm was a 1-s train (20 Hz) applied for slightly less than one-half the movement cycle. The rationale for this paradigm was to activate the muscles during specific periods of the step cycle, more-or-less as they might be activated during locomotion. More importantly we wanted to avoid muscle fatigue over the 10-15 movement cycles we recorded. This intermittent stimulus paradigm with 1 s of stimulation followed by a 1.8-s rest did not appear to induce fatigue because the results were highly reproducible across cycles.
We applied the stimulus train to one muscle group at a time. One train (stimulus 1) was applied at the onset of the forward trajectory of the cycle and to the other (stimulus 2) at the onset of the backward trajectory. The muscles for which each stimulus produced a maximal response are summarized in Table 1 for each of the four cats used for these experiments. As we note in the preceding text, however, the stimuli were not specific to one muscle although the effects seemed to be largely confined to synergists.
|
Kinematic measurements.
We placed reflective markers (6 mm in diameter) on the skin over the hip, knee, ankle, and lateral metatarsal-phalangeal joint of the foot and corrected the digitized positions of the markers in the image plane for skin slippage at the knee and also for out-of-plane positions using the algorithm described in detail in Eian and Poppele (2002). The kinematic data taken at 60 frames/s were resampled at 30/s to correspond to the 33.3-ms bin width of the neural data (see following text).
We represented limb kinematics in the coordinates of the limb axis and the anatomical joint angles. The limb axis is the segment joining the hip joint and the foot, and it defines foot position in polar coordinates by its orientation angle O, measured clockwise from the horizontal to the axis, and its length L, in centimeters (Fig. 1B). The joint angles are also defined in Fig. 1B as the angles measured clockwise between limb segments (see also Bosco et al. 1996).
Neural activity
We recorded DSCT unit activity from their axons in the dorsolateral funiculus at the T10-T12 level of the spinal cord using insulated tungsten electrodes (5 M
, FHC, Brunswick, ME). Units were identified as spinocerebellar by antidromic activation from the white matter of the cerebellum and/or from the restiform body. Activity recorded continuously during series of 10-15 passive step-like movement cycles was aligned to the reference starting point in each cycle at the beginning of the forward swing (black circle in the foot trajectory in Fig. 1A). We computed cycle histograms using the aligned, binned activity (bin width: 33.3 ms) from five to seven consecutive cycles (see Bosco et al. 2003).
Data analysis
KINEMATIC DATA. We expressed each joint angle trajectory as the difference from the mean angle over the movement cycle and defined each point in the trajectory by the three joint angles plotted in a three-dimensional (3D) representation of the movement cycle kinematics. This is similar to the type of representation used to describe the limb kinematics of passive limb postures (Bosco et al. 1996), the kinematics of stance in behaving cats (Lacquaniti and Maioli 1994), and the kinematics of walking in humans (Grasso et al. 1998, 2000). In all these examples, the data points representing each foot position fell on or near a plane that explained a large fraction of variance in the data set. They were not scattered throughout the 3D joint space as expected if joint angles varied independently. Based on these findings, we also quantified the limb kinematics of the control and perturbed steps by fitting least-square planes to the joint angle trajectory data. The planes were defined by the direction cosines of the vector normal to the plane and the fraction of the variance explained by each plane.
Because the foot was attached to the robot platform, the limb axis trajectory was assumed to be identical under all conditions. In some cases, however, the limb axis trajectory appeared to differ slightly across conditions. This was due to some relative motion between the foot marker and the robot during the step cycle as the limb was alternately supported on the toes and on the flat foot. The marker on the metatarsal-phalangeal joint, which was slightly proximal to the footpad attachment to the robot, was sometimes displaced along the limb axis relative to the footpad during a movement cycle. The result was that the foot marker may not have had exactly the same trajectory in the control and perturbed conditions for example. However, the limb length measured to the footpad or to the robot platform was not affected by the constraint nor were the measurements of limb axis orientation.
NEURONAL DATA. For each DSCT neuron, we determined whether movement-related activities differed significantly between the control and perturbed conditions by applying the two-sample nonparametric Kolmogorov-Smirnov (K-S) test to pairs of cycle histograms recorded in the two conditions. Differences were considered statistically significant at the 0.05 level. We also evaluated the possibility that activity differences between control and perturbed steps might reflect inherent firing variability by applying the same test to pairs of activity histograms determined from separate trials recorded in a given condition.
We evaluated the population behavior of the neurons by determining the average changes in activity produced by the joint constraint or VRS and also by applying a principal component analysis (PCA). We determined the principle component (PC) waveforms for the total set of neurons studied in the control condition with no constraint or stimulation. We then used the response waveforms that were found to be different from the control for each experimental condition for separate PCAs (see Bosco and Poppele 2002; Bosco et al. 2003 for further details).
All statistical analyses were performed using SYSTAT (Wilkinson, 1990).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
; Poppele et al. 2002). The foot path trajectory and therefore the limb axis trajectory was basically the same for all recordings (see Fig. 1). Each cell was recorded first in a control condition where the limb motion was determined only by its passive biomechanics, and then it was recorded in one or more test conditions. We applied a joint angle constraint that modified the limb geometry as a test condition for 67 cells and electrical VRS for a portion of each movement cycle as the test condition for 51 cells. Both forms of perturbation were applied while recording from 29 of the 89 cells. Passive movement kinematics.
The kinematics of the joint angle trajectories during a movement cycle are illustrated for the control and joint constraint conditions in Fig. 2. The graphs show the angular trajectories of the hip, knee, and ankle angles as a function of time over a cycle averaged for the six animals used in this study. The SDs of the control trajectories illustrate the small variance we observed among these animals. The trajectories for each animal were also highly reproducible across trials because they were determined by the passive mechanics of the system and the motion of the robot. During the joint constraints, it is evident that the knee constraint effectively blocked the motion of the knee joint to <5° on average (gray traces). In contrast, the elastic constraint at the ankle was not very effective in limiting ankle joint motion; however because it limited the changes in shank elevation relative to the robot platform, it did alter significantly the angular trajectories of the hip and knee (gray dashed traces).
|
Muscle stimulation also tended to change the limb geometry for at least the duration of the stimulation. The effects were somewhat different for each animal though, no doubt reflecting the different muscles involved (see Table 1). Figure 3 shows the effects of the VRS in four animals (colored traces) compared with the controls (black traces). The largest effects of VRS were noted for cat 2, which are plotted with thicker lines. The changes were mostly confined to the stimulus periods as expected. Stimulus 1 had little or no effect on the joint angle trajectories, but it produced a large change in the knee angle trajectory for cat 2 (Fig. 3, red traces). Stimulus 2 was generally more effective in altering joint angle trajectories than stimulus 1 (Fig. 3, blue traces), and again the changes tended to be greater for cat 2.
|
We quantified the effect of the perturbations on the joint angle kinematics by fitting least-square planes to the values of the three angles during the movement (Fig. 4). The same type of planar relationships seen in cats and human subjects during normal behavior also provided an adequate description of cat hindlimb joint-angle kinematics during the passive step-like movements. In fact, 73% of the total joint angle variance (81.5 ± 5.9%, mean ± SD; see Table 2) was explained by the joint angle covariance planes, and these relationships were roughly invariant across cats. The control graphs in Fig. 4 illustrate the similarity of covariance plane orientations for two cats, and this is also evident in the similar values of their direction cosines for all six animals1 (see Table 2).
|
The constraints were designed primarily to alter limb geometry by changing the covariation of joint angles for a given set of limb endpoint locations. The large differences we observed in the covariance plane orientations illustrate the extent to which this was accomplished (Fig. 4, Table 2). The covariance planes also explained a smaller fraction of variance when the knee joint was constrained (mean: 67% ±9.4, Table 2), signifying a weaker coupling among joint angles. The coupling was actually increased though for the two cases in which we applied the ankle constraint, and this constraint had a minimal effect on the joint angle covariance.
The effect of VRS on the joint angle covariance was also variable, but it was generally less than the effect of the joint constraints. Both stimuli induced only minor rotations of the covariance plane from the control (Fig. 4A). However, there appeared to be a significant reduction in joint coupling because the linear covariance accounted for <60% of the total variance (58 ± 13%). In fact, a linear covariance was generally inadequate to account for the entire trajectory because the covariance tended to be different during the stimulus and no-stimulus periods (e.g., Fig. 4A, stimulus 1). A separate regression of joint angles during just the stimulus period showed that the stimuli could increase or decrease the linear joint angle coupling (Table 2, *). For cats 1, 3, and 4, for example, stimulus 1 increased the coupling over the control as indicated by the greater variance explained by the linear relationship. In contrast, stimulus 2 decreased the strength of the linear coupling for cats 3 and 4.
In summary, both types of perturbation, the constraints and the VRS had significant effects on the knee angle trajectories, and they had mixed effects on the joint angle coupling.
DSCT unit responses
We examined the responses of 67 DSCT neurons to these step-like movements with joint constraints (50 with knee and 21 with ankle constraint including 4 cells recorded under both constraint conditions), and 51 neurons during VRS. For each cell, we determined two response histograms (5-7 cycles each) for each condition to allow a comparison between inherent firing variability and differences due to the experimental condition. In general there was a very low level of intrinsic variability such that differences across trials did not reach a level of statistical significance for any of the cells studied (2-sample K-S statistics, P < 0.05) (see also Bosco and Poppele 1999)
Effect of joint constraints
The constraints had variable effects on the DSCT responses, including no discernible effect on the responses of 34 neurons (51%; K-S test P > 0.05; see Fig. 5). The other 33 cells (49%) were all affected to some extent, as illustrated by examples in Figs. 6 and 7. In general, the response activity of the affected cells did not appear to follow a common pattern, and the effect of the constraint on unit responses seemed to occur in different parts of the cycle (e.g., cells 2680 and 2730 in Fig. 7). It is interesting to note that the few cells we did record under both constraint conditions in cat 5 exhibited similar changes in their responses to both types of constraint (cells 2589 and 2602 in Fig. 6).
|
|
|
There were also changes in amplitude in 73% of the cells with altered responses, 22% were decreased, and 51% increased by 5 imp/s.
Effect of muscle stimulation
The VRS generally elicited short-latency excitatory or inhibitory responses from the DSCT neurons (Bosco and Poppele 1997), but stimulating the same muscles during movement did not always alter their responses in the same way. Stimulus 1 altered the movement responses of only about half of the cells (54% or 25/46 tested), and stimulus 2 altered the responses of another half (48%, 21/44 tested). Only 20% of the cells (8/40 tested) were significantly affected by both stimuli. Overall 73% (37/51) of the cells responded to the passive movement differently when one or the other stimuli was active. However, we found that 95% of the 37 cells tested with VRS applied when the limb was not moving responded significantly to both stimuli.
In most cases, the stimulus effect occurred only during the stimulus period with some rebound effect immediately after the stimulus. The response behavior illustrated in Fig. 7 is typical of what we observed. The responses of cells 2680 and 2675 were both affected by stimulus 1 and not stimulus 2, whereas the responses of cell 2728 and 2730 were only affected by stimulus 2. Note that both stimuli exerted clear responses when applied in a random sequence when the limb was not moving. While some random stimulus responses were consistent with the effects observed during movement (e.g., stimulus 1 for cell 2680 and stimulus 2 for cell 2730), mostly they were not. For example, cell 2675 was inhibited by both random stimuli when the limb was not moving, although the inhibition was preceded in both cases by a short-latency excitation. During the movement, however, stimulus 1 enhanced the response and stimulus 2 had no apparent effect.
The examples illustrated in Fig. 7 also show the effect of the knee constraint on these cells (gray curves). There were significant effects on the responses of three of them (cells 2680, 2675, and 2736) and no effect on the other (cell 2728). In general, there was no relationship between whether a cell's response was altered by the constraint or by the VRS.
Population behavior
The average differences evoked by knee constraints across cells were fairly consistent because they mostly involved response differences during the latter part of the forward trajectory with little or no change during the backward trajectory (Fig. 8A). The changes evoked by the ankle constraint seemed more variable though, having their major differences on average during the beginning of the forward trajectory and/or the end of the backward trajectory (Fig. 8B)
|
The average differences in the responses evoked by VRS occurred primarily during the stimulus periods. When we examined the increases and decreases separately for the two stimuli, they seemed to be mostly step-like changes in activity added to the response seen without stimulation (Fig. 8, C-F). There was little evidence for a consistent dynamic effect except at the onset or offset of the stimulus. Stimulus 1 tended to evoke a slow onset and stimulus 2 more often evoked a rapid onset with an overshoot.
We also examined the population behavior by means of a PCA to identify common features in the response waveforms. The waveform of a response histogram can be described as a weighted sum of constituent factors or PCs that are common to the waveforms in a set. The analysis of the total set of control responses waveforms produced three PCs that accounted for 81% of the total variance (Fig. 9, black traces). Higher-order PCs were not considered significant (eigenvalues <1.0) and each explained <3% of the variance.
|
A PCA of the 33 responses that were changed by the constraints gave basically the same PCs (Fig. 9, gray traces). The first four PCs were nearly identical and together accounted for 84 and 87% of the total variance in the control and test responses, respectively (see also Bosco and Poppele 2002). In fact, the PCs obtained from the responses altered by the constraints accounted for 97% percent of the respective waveform variance in the control PC 1 waveform and 87% of the variance in the control PC 2 waveform. This implies that the changes in response waveforms produced by either type of constraint were not due to new or missing response components. Therefore the control and test responses may all be reconstructed from a weighted linear sum of the same PCs, and the differences in response waveform can be accounted for by different weightings. Note that the variance explained by each PC was somewhat different for the changed responses (Fig. 9B). In particular, the contribution of the second PC was 50% greater (18 vs. 12%) for the changed responses, although the cumulative variance explained by PC1 and PC2 was the same in both conditions (73%)
The PCA gave a somewhat different result for the muscle stimulation (Fig. 10). In this case, the first and third PCs were unchanged in either stimulus condition, but the second PC waveform was altered. Thus while the PC1 obtained from the responses altered by the VRS accounted for between 97 and 99% of the waveform variance in the control PC1 waveform, the PC 2s accounted for between 29 and 34% of the variance in the control PC 2 waveform. Furthermore, additional PCs became significant with both stimuli, and these appeared to reflect stimulus effects. For example, a new PC obtained from responses altered by each stimulus (Fig. 10A, PC 4) reflected a transient change in the activity waveform at the beginning or end of the stimulus period. Thus it appears that some of the stimulus effects were expressed independently from the basic kinematic response components (i.e., they were represented by separate PCs) and others were not (i.e., they contributed to a change in the 2nd PC waveform).
|
To control for the heterogeneity of effects due to stimulating different sets of muscles in each animal, we did a separate PCA using only the response data from cat 2 (Fig. 10C). We recorded from 27 cells in this cat during the VRS and observed changes in the responses of 18 to one or both stimuli. The VRS had clear effects on both the joint angle trajectories (particularly the knee) and also on the joint coupling in this animal. The result of the PCA, plotted in Fig. 10C, was that the first two PCs accounted for between 65 and 80% of the waveform variance in this subset, and the PC waveforms were basically the same as for the total data set. The differences between the control PC 2 and the PC 2's determined in the different stimulus conditions were similar for this cat and for the total data set, even though the kinematic effects of the stimuli seemed to be greatest for this cat. Thus it does not appear that the responses are simply driven by joint angle trajectories.
Relationship to joint angle trajectories
Because the second PC does seem to resemble the knee angle trajectory, we also examined the extent to which the response changes were correlated with the changes in the joint trajectory. The knee angle trajectory was the one most changed by both procedures. An inspection of Figs. 8A and 2 (knee trajectories) does suggest a similarity between the average difference in response and the knee trajectory (or change in knee trajectory where the knee angle was fixed by the constraint). However, the average response change is monotonic during the forward trajectory and constant or absent during the backward movement while the knee trajectory undergoes both flexion and extension in the forward trajectory and large extension during the back trajectory. Correspondingly the correlation between the average response change and knee trajectory change is rather weak (r = 0.48, range: 0.006 < r < 0.85 for correlations done cell by cell).
Overall we found that the control responses were also poorly correlated with the knee trajectory overall (average: r = 0.29), and those correlations were different during VRS. The correlation increased (0.32) with stimulus 1 for cat 2 and decreased with stimulus 2 for cats 3 (0.07) and 4 (0.23). Therefore the relationship between response waveform and knee angle trajectory was different in the control and stimulated states, implying that the joint trajectory did not determine the response change. The correlation between the average response change and the knee trajectory was also low for the VRS effects (r = 0.44, cell-by-cell range: 0.01 < r < 0.85)
Component weighting
The data presented in the preceding text suggest that the changes we observed in the movement responses may be accounted for by the re-weighting of an invariant set of components that represent the limb axis kinematics. To examine this issue further, we determined the weighing coefficients for the first two PCs and compared them in the control and experimental conditions. Figure 11 shows the weighting coefficients for the first PC plotted against that of the second PC for the joint constraint data. A connecting line tracks the behavior of each affected cell from the control (
) to the constrained response coefficients (
, knee; 
, ankle constraint). The lines are mostly vertical indicating predominant changes in the weighting of the second PC rather than the first (see also Bosco and Poppele 2002), and there are no obvious or systematic differences between the changes evoked by the two types of constraint. This result is further quantified in 11B by the circular distribution of the direction angles of the vectors connecting control and constrained responses. This distribution has a highly significant bimodal trend along an axis oriented almost vertically (Raleigh test, P < 0.01; major axis: 85°) that is, parallel to the second PC axis.
|
A similar analysis for the muscle stimulation data could not be done because the addition of higher-order PCs changed the total percentage of variance explained by PC 1 and PC 2 (Fig. 10). Moreover, the PC 2 itself was altered by the stimulus.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
), whereas those altered by the VRS could be accounted for by new or modified response components related directly to the VRS.
One possible interpretation of these results is that the altered responses simply reflect the local changes in sensory input that accompany the perturbations of limb mechanics. According to this interpretation, response waveform changes would result from modifications of response components related directly to the joint angles that were constrained or stiffened by muscle contraction. For example, response components related to the knee and ankle angle trajectories would be expected to be attenuated when constraints where applied because they reduced the angular excursions of the knee and the ankle. Likewise, the response components related to the hip angle trajectory would not be affected because of the minimal affect of the constraints on this trajectory. The differential effect on the weightings of PC 1 and PC 2 seems consistent with this interpretation. PC 1 correlates about as well with hip angle trajectory as is does with the limb axis orientation trajectory, and PC 2 also correlates with the knee and ankle trajectories (Poppele et al. 2002), although less well than it correlates with the limb axis length trajectory. However, it is important to note that the joint constraints did not simply attenuate the knee angle trajectories, they also modified their waveforms. Thus if there was a direct relationship between the PC 2 waveform and the joint angle trajectories, it would have led to an altered waveform in the constraint data. Instead the PC 2 waveform was invariant and continued to correlate with the limb axis length trajectory, which was not changed by the constraint. Furthermore we did not find a correspondence between the average changes in the responses and the changes in the joint angle trajectories.
This issue is less straightforward though for the VRS because the stimulus was not applied continuously during the movement. In this case, the waveform of PC 2 was altered by the stimulation, and the changes were much more evident during stimulus 2, which also seemed more effective in altering joint angle trajectories. The effect was not consistent across experiments, however, because different sets of muscles were stimulated in each experiment. In fact, the PCs derived from a separate analysis of the cat 2 responses showed about the same differences between control and stimulus conditions as did the PCs derived for the entire data set. Yet cat 2 exhibited the largest changes in joint angle trajectories and one of the larger changes in joint angle covariation. Thus a clear relationship between the changes in PC 2 and the joint angle trajectories is also not evident in the VRS data.
Based on such considerations, we propose that the altered responses do not simply reflect the sensory consequences of the biomechanical perturbations but rather that they result from modulatory changes occurring within the spinal circuitry. This interpretation gains some support from the results of the accompanying paper (Bosco et al. 2003), which shows that serotonergic modulation can selectively affect specific response components, suggesting that components revealed by the PCA can be independently modulated within the spinal circuitry.
We propose that they may also be modulated in a similar way by the sensory input (Fig. 12). With the joints constrained, the altered sensory input seems to alter primarily the gain of the PC 2 (length) component of the DSCT responses. With the VRS, we observed changes in both the PC 2 component waveform and in the overall waveform of the DSCT responses. In this case, the sensory input generated by the VRS appears to have both a presynaptic and postsynaptic effect.
|
Sensory modulation of sensory pathways is certainly not a new concept, and a number of mechanisms have been proposed (e.g., Melzak and Wall 1965; Rudomin 1999a; Sillar 1991). Presynaptic inhibition of primary afferent pathways resulting from sensory stimulation is one well-studied example of the sensory modulation of sensory input (Rudomin 1990). Although the functional role of this mechanism has been examined at the level of segmental afferents, not much is known about its possible role in the central transmission of proprioceptive information. However, Rudomin and colleagues (Lomeli et al. 1998; Rudomin 1999b) showed that presynaptic inhibition of primary afferent synapses can be controlled independently for segmental and spinocerebellar pathways. The functional consequences of such mechanisms have not been systematically explored, however.
We believe the results we obtained with both the joint constraints and VRS are consistent with a sensory modulation that occurs presynaptically to the DSCT neurons. The presence of sensory modulation seemed particularly evident in its effect on the VRS responses. Nearly all the cells responded to both stimulus 1 and stimulus 2 applied when the limb was not moving. However, only 20% of the cells responded to both stimuli during limb movement. Because the two stimuli were applied in different parts of the trajectory, this result suggests that the foot position and direction of movement determined, in some way, the presence of a response. The movement may have also affected the type of response as well because we could not always predict from the still-limb response whether the VRS would produce an increase or decrease in activity during movement. Thus the responses recorded during movement were not a simple summation of a basic movement response and the VRS response observed with the still limb.
The presynaptic basis of the sensory modulation is evident in the effect of the constraints. Those changes consisted of significant alterations in the response waveforms, and the PCA demonstrated that the changes could be accounted for by re-weighting or modulating the underlying response components present in the control responses. Thus it seems most likely that the sensory input altered the gain or sensitivity of specific inputs to DSCT neurons, which were not themselves altered by the constraints. That is, the underlying response components observed in the control responses were not affected but the efficacy of their transmission to the DSCT cells was.
This was not the case for the VRS, which altered the control waveform of PC 2 and also introduced new, higher-order PCs. Thus the VRS effect on the responses was not independent from the waveforms of the underlying components as the constraint effect was. However, the addition of new components suggests that a part of the effect may also have resulted from a new input to the DSCT. This might represent a direct input to DSCT neurons from tendon organ receptors that were activated by the muscle stimulation (Osborn and Poppele 1983). In contrast, the changes in the PC 2 waveform are more consistent with an effect on response components having a presynaptic origin involving some interaction between an underlying movement response and the VRS. Moreover it was the response component associated with PC 2 that was altered and not that associated with PC 1 or PC 3. That is, the VRS appears to have altered one component of the movement response in addition to altering its transmission efficacy. These effects of the VRS on the PCA results also serve as a kind of control for both the constraint results and the drug results reported in Bosco et al. (2003). The VRS showed that when new response components are introduced by the stimulus, they do indeed give rise to new PCs not present in the controls, and/or they alter control PC waveforms.
The VRS, however, was not applied over the entire trajectory, so we cannot rule out that in this case the altered movement responses were simply due to a change in sensitivity occurring only during the stimulus. Unfortunately we were not able to test this further by applying the VRS over the entire cycle because the fatigue that occurs during continuous stimulation at 20 Hz would have made the interpretation of results much more difficult. Moreover because we did not monitor the sensory input from the limb, we cannot rule out that the contractions were simply more effective in activating the appropriate receptors in some limb positions and not in others. This is not entirely unlikely because we selected the two ventral rootlets for stimulation according to their ability to activate generally antagonistic sets of muscles.
The VRS result does seem to be compatible though with our previous report that VRS applied during static postures only altered the gain or sensitivity without changing a cell's preferred direction (Bosco and Poppele 2000). Correspondingly, the VRS during movement appeared to simply enhance or depress the activity uniformly during the stimulus period with respect to the baseline (control). Although we did note some dynamics associated with these changes, they may reflect the way in which the sensory receptors responded to the onset and offset of the contractions. Thus we would predict that if we were able to stimulate continuously throughout the movement cycle, we would not have seen changes in PC response waveforms but rather weighting changes reflecting changes in gain of one or more response components.
It is striking that both forms of compliance perturbation had their effect on PC 2 with no effect on the other major response components. It suggests to us that perhaps PC 1 and PC 2 may represent different kinds of information that correlate with limb kinematics, even though they may not all represent kinematics directly. For example, if PC 2 was associated with some aspect of limb compliance or stiffness, it might still be correlated with whole-limb limb kinematics in some way, yet that association would vary with changes in stiffness. Limb stiffness tends to be maximal along the axis of the limb (the length dimension) and minimal along a perpendicular axis (Mussa-Ivaldi et al. 1985). This relationship might account for why procedures that modify limb stiffness independently from limb axis length can also alter the weighting of PC 2. The reason that some cells are affected more than others might reflect some kind of distributed representation of local limb compliance across cells.
Whatever the actual association between response components and limb biomechanics might be, our analysis of this system has shown that DSCT responses have two major independent components and they can be modulated separately. We also showed a clear correlation between the waveforms of those components and the kinematics trajectories of the passive limb axis. These findings are not incompatible with alternate interpretations about what information these response components may provide to the nervous system. The idea that limb stiffness or compliance may be explicitly represented could have interesting implications about cerebellar function (Bosco and Poppele 2001). For example, it might indicate that the cerebellum monitors limb stiffness through this spinocerebellar system and then plays a role in regulating limb stiffness thorough its influence on reflex pathways.
|
DISCLOSURES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
|
FOOTNOTES |
|---|
1 Because we had previously analyzed the joint angle covariance planes during passive hind foot placement for some of the same cats used in this study (cats 1-4, Bosco et al. 2000), we also compared the two sets of results. Overall, the coupling among joint angles may have been somewhat stronger for the postures than during the movements as indicated by a slightly greater fraction of the variance explained by the covariance planes (85.7 ± 3.6%; mean ± SD). The somewhat larger unexplained variance during movement may have resulted from small but significant local nonlinearities with respect to the best fitting plane that were most evident during the forward trajectory where the footpath velocity was generally highest (Fig. 1C). Apart from these relatively small differences, the orientations of the covariance planes during passive movements and static postures were rather similar, the average difference being only 6.1 ± 5.8° (mean ± SD). 
Address for reprint requests and other correspondence: R. E. Poppele, Dept. of Neuroscience 6-145 JH, 321 Church St. SE, Minneapolis, MN 55455 (E-mail: dick{at}umn.edu).
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
Bosco G and Poppele RE. Low sensitivity of dorsal spinocerebellar neurons to limb movement speed. Exp Brain Res 125: 313-322, 1999.[ISI][Medline]
Bosco G and Poppele RE. Reference frames for spinal proprioception: kinematics based or kinetics based? J Neurophysiol 83: 2931-2945, 2000.
Bosco G and Poppele RE. A spinocerebellar perspective on proprioception. Physiol Rev 18: 539-568, 2001.
Bosco G and Poppele RE. Encoding of hindlimb kinematics by spinocerebellar circuitry. Arch Ital Biol 140: 185-192, 2002.[ISI][Medline]
Bosco G, Poppele RE, and Eian J. Reference frames for spinal proprioception: limb endpoint based or joint-level based? J Neurophysiol 83: 2946-2955, 2000.
Bosco G, Rankin A, and Poppele RE. Representation of passive hindlimb postures in cat spinocerebellar activity. J Neurophysiol 76: 715-726, 1996.
Bosco G, Rankin A, and Poppele RE. Modulation of dorsal spinocerebellar responses to limb movement. I. Effect of serotonin. 90: 3361-3371, 2003.
Eian J and Poppele RE. A single-camera method for three-dimensional video imaging. J Neurosci Method 20: 65-83, 2002.
Grasso R, Bianchi L, and Lacquaniti F. Motor patterns for human gait: backward versus forward locomotion. J Neurophysiol 80: 1868-1885, 1998.
Grasso R, Zago M, and Lacquaniti F. Interactions between posture and locomotion: motor patterns in humans walking with bent posture versus erect posture. J Neurophysiol 83: 288-300, 2000.
Lacquaniti F and Maioli C. Independent control of limb position and contact forces in cat posture. J Neurophysiol 72: 1476-1495, 1994.
Lomeli J, Quevedo J, Linares P, and Rudomin P. Local control of information flow in segmental and ascending collaterals of single afferents. Nature 395: 600-604, 1998.[Medline]
Melzak R and Wall PD. Pain mechanisms: a new theory. Science 150: 971-979, 1965.
Mussa-Ivaldi FA, Hogan N, and Bizzi E. Neural, mechanical, and geometric factors subserving arm posture in humans. J Neurosci 5: 2732-2743, 1985.[Abstract]
Osborn CE and Poppele RE. Cross-correlation analysis of the responses of units of the dorsal spinocerebellar tract (DSCT) to muscle stretches and contractions. Brain Res 280: 339-342, 1983.[ISI][Medline]
Poppele RE, Bosco G, and Rankin A. Independent representations of limb axis length and orientation in spinocerebellar response components. J Neurophysiol 87: 409-422, 2002.
Rudomin P. Presynaptic inhibition of muscle spindle and tendon organ afferents in the mammalian spinal cord. Trends Neurosci 13: 499-505, 1990.[ISI][Medline]
Rudomin P. Presynaptic selection of afferent inflow in the spinal cord. J Physiol 93: 329-347, 1999a.
Rudomin P. Selectivity of presynaptic inhibition: a mechanism for independent control of information flow through individual collaterals of single muscle spindle afferents. Prog Brain Res 123: 109-117, 1999b.[ISI][Medline]
Sillar KT. Spinal pattern generation and sensory gating mechanisms. Curr Opin Neurobiol 1: 583-589, 1991.[Medline]
Wilkinson L. The System for Statistics Evanston, IL: SYSTAT, 1990.
This article has been cited by other articles:
|
|
 |
|
  M. L. Willson, C. McElnea, J. Mariani, A. M. Lohof, and R. M. Sherrard BDNF increases homotypic olivocerebellar reinnervation and associated fine motor and cognitive skill Brain, April 1, 2008; 131(4): 1099 - 1112. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. B Stein, D. J Weber, Y Aoyagi, A Prochazka, J. B. M Wagenaar, S Shoham, and R. A Normann Coding of position by simultaneously recorded sensory neurones in the cat dorsal root ganglion J. Physiol., November 1, 2004; 560(3): 883 - 896. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||
