| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
The Journal of Neurophysiology Vol. 80 No. 1 July 1998, pp. 59-70
Copyright ©1998 by the American Physiological Society
1 Department of Physiology, Northwestern University Medical School, Chicago, Illinois 60611; and 2 Center for Sensory-Motor Interaction, Aalborg University, DK-9220 Aalborg, Denmark
 |
ABSTRACT |
|---|
 Abstract
 Introduction
Methods
Results
Discussion
References
|
|---|
Miller, L. E. and T. Sinkjaer. Primate red nucleus discharge encodes the dynamics of limb muscle activity. J. Neurophysiol. 80: 59-70, 1998. We studied the dynamical relationship between magnocellular red nucleus (RNm) discharge and electromyographic (EMG) activity of 10-15 limb muscles in two monkeys during voluntary limb movement. Recordings were made from 158 neurons during two different kinds of limb movement tasks. One was a tracking task in which the subjects were required to acquire targets displayed on an oscilloscope by rotating one of six different single degree of freedom manipulanda. During this task, we recorded the angular position of the manipulandum. The monkeys also were trained in several free-form food-retrieval tasks that were much less constrained mechanically. There was generally significantly greater neuronal discharge during the free-form tasks than during the tracking task. During both types of tasks, cross-correlation and impulse response functions calculated between RNm and EMG were predominantly pulse-shaped, indicating that the dynamics of the RNm discharge were very similar to those of the muscle activity. There was no evidence during either task for a substantial dynamical transformation (e.g., integration) between the two signals as had been previously suggested. In only 15% of the cases, did these correlations have step or pulse-step dynamics. There was a relatively broad, unimodal distribution of lag times between RNm and EMG, based on the time of occurrence of the peak correlation. During tracking, the mode of this distribution was ~50 ms, with 80% of the lags falling between 
100 and 200 ms. During the free-form task, the mode was between 0 and 20 ms, with 65% of the lags between 100 and 200 ms. A positive lag indicates that RNm discharge preceded EMG. The shape and timing of both the cross-correlation and the impulse response functions were consistent with a model in which many RNm neurons contribute mutually correlated signals which are simply summed within the spinal cord to produce a muscle activation signal.
One means by which we gain insight to the organization of the motor control areas of the brain is to measure the activity of individual neurons and then to relate the modulation in discharge rate to various aspects of the animal's motor behavior. Typically, ensemble averages of discharge rate are constructed with respect to particular behavioral events (Crutcher and Alexander 1990 We recorded EMG and neuronal data from two Macaca mulatta monkeys (AL and BR). All animal care, surgical, and research procedures were approved by the Institutional Animal Care and Use Committee of Northwestern University. During training and experiments, the monkeys received most of their water and fruit as reinforcement for desired behavior. The monkeys' weight and the volume of fluid consumption was checked daily. At 3- to 4-wk intervals, they received free water for a weekend. The monkeys were seated in a primate chair that allowed nearly unrestricted movements of the right arm and were trained to perform two basic types of movement tasks. In the free-form tasks, the monkey retrieved a small piece of food from the experimenter's hand, a Kluver board, or from behind a clear plastic barrier (Miller et al. 1993) Surgical procedures
After training, aseptic surgery was performed under halothane/nitrous oxide anesthesia. The monkeys received dihydrostreptamycin plus penicillin G (Combiotic) antibiotic immediately before surgery and for 3-5 days after surgery. Intramuscular injections of buprenorphine hydrochloride (Buprenex), a long-lasting analgesic, also were given for several days.
Neuronal recordings and histological localization
Neuronal recordings were made with epoxy-coated tungsten microelectrodes having an impedance between 0.25 and 1 M Data collection and analysis
The microelectrode signals were amplified, filtered, and passed through a level discriminator. The discriminator pulses were low-pass filtered (10-ms time constant; 15 Hz) to produce a signal proportional to the firing rate of the cell. EMG signals were amplified, rectified, low-pass filtered, and sampled at 200 Hz. The same filter applied to both neuronal and EMG signals prevented phase shifts of one signal relative to an other. We could collect only 16 analogue signals, hence it was generally not possible to collect all EMG signals. We also were unable to collect kinematic data during the free-form movements. Data files of 1- to 2-min duration were collected while the monkey performed a single task. For a given unit, we attempted to record at least one data file for each behavioral task. Because there was no assurance that the cell could be held long enough to record files for all tracking devices, we typically began recording first with those tasks that appeared to evoke the most reliable responses.
RNm discharge during tracking and free-form limb movements
To study the modulation of the control signals represented by the discharge of single RNm neurons, we used two different types of limb movement tasks. During the free-form tasks, all of the implanted muscles became active in some phase of the movement with a wide variety of patterns of activation across muscles. The complexity of these patterns was important as it allowed us to discriminate among the muscles involved in the task, those the discharge of which bore greatest resemblance to that of a given neuron. In contrast, during the tracking task, activity was largely confined to brief bursts of a few muscles. Although this task was not as well suited for the analysis of spatial properties of RNm discharge, it was ideal for studying the timing and dynamics of individual, well-correlated neuron/muscle pairs.
RNm/EMG dynamics
Ideally, an impulse response function, rather then a cross-correlation, would be used to measure the dynamical transformation between signals (Houk et al. 1987)
RNm/EMG timing
The time of occurrence of the peak correlation provides a measure of the mean delay between the neuronal signal and EMG for the entire duration of the signals rather than only at the onset of movement. When the peak was sharp, as in Fig. 5 for EDC, or lumbrical in Fig. 7, the estimate could be very precise. In these examples, the EMG followed the neuronal signal with delays of 50 and 75 ms, respectively. In other cases, when the peak was quite broad, a precise determination of delay was less meaningful.
Simulation of RNm/EMG dynamics
The great majority of these lags under all conditions were positive, meaning that RNm discharge led correlated muscle activity. However, there were a significant minority of cases, particularly during the free-form behavior, in which RNm discharge followed muscle activity. These observations must be explained by some means in addition to a direct connection from RNm neuron to muscle. One plausible explanation is that the negative lag correlations resulted indirectly from correlated input sources common to both the neuron and muscles. During the tracking task, the latency between the onset of discharge and movement covered a range from 40 to 160 ms for well-correlated neurons. Others have made similar observations (Gibson et al. 1985a
Our results have demonstrated that the time course of RNm discharge closely resemble those of well-correlated limb muscles in both unconstrained free-form limb movements and a single degree of freedom tracking task. Very little dynamical transformation occurred between RNm cells and muscles. The tracking task, in particular, revealed a delay, measured by the peak of the correlation, averaging 50-75 ms. During free-form movements, lags were distributed more broadly across the population of units, including negative lags and long positive lags. We have suggested that the shape of the cross-correlation function and the broad range of lag times result from substantial correlations in discharge rates among the population of RNm neurons. The nature of this correlation appears to be a critical factor determining the dynamics of the descending motor command.
RNm dynamics
There is basic agreement that the discharge of RNm cells is primarily phasic. Whether there is any tonic component of the RNm discharge or whether the discharge is better related to kinematic or dynamic variables (Ghez and Vicario 1978 EMG dynamics
We have shown previously that many more correlations occurred between RNm discharge and distal muscles, particularly the digit extensors, than for the more proximal muscles (Miller and Houk 1995) Spinal cord dynamics
Because of the relation to velocity, Gibson and colleagues (1985b) suggested that an integrator in the spinal cord might be required to convert the phasic RNm signal into a tonic muscle activation signal. The pulselike cross-correlations and impulse responses calculated between RNm units and muscle activity indicate that there was generally not a large dynamical transformation between these signals. Only occasionally was an RNm/EMG pair found with integrator or pulse-step-type dynamics. In other words, most of the pulselike RNm bursts were "integrated" by the mechanics of the limb rather than by a neuronal integrator in the spinal cord.
Timing of the descending motor command signal
For both red nucleus and motor cortex, the conduction delay of individual action potentials measured by spike triggered averaging, between 5 and 10 ms, (Fetz and Cheney 1980 Dynamics of the descending motor command signal
We formulated a model in which muscle activity resulted from the summed activity of many premotor (e.g., RNm and M1) neurons, each of which had activity correlated with but significantly delayed or advanced from that of the measured neuron. Hence, the timing of the peak RNm/EMG cross-correlation could be thought of as a measure of the timing of the recorded RNm unit against that of the entire population of correlated neurons. The model suggested that much of the width of the cross-correlation and impulse response functions resulted because the estimate of the dynamics between a measured RNm neuron and muscle activity was biased by the presence of input noise. For simplicity, the shifts between the input neuron and each of the correlated neurons were drawn from a normal distribution. This choice largely determined the shape of the resultant impulse response function. For example, the model accurately predicted that the more tightly clustered range of lags during tracking would result in narrower correlation peaks compared with those of the free-form task.
INTRODUCTION
Abstract
Introduction
Methods
Results
Discussion
References
; Evarts and Tanji 1976; Georgopoulos et al. 1982; Thach 1978). Alternatively, mean discharge rate during movement may be calculated and correlated against measures of motor behavior (Cheney and Fetz 1980; Gibson et al. 1985b; Lamarre et al. 1983). Both methods require that the animal repeatedly perform a stereotypic movement.
; Gibson et al. 1985b). Gibson and colleagues speculated that an integrator, located in the spinal cord, might transform the brief bursts of neuronal activity into tonically maintained muscle commands. However, such a transformation would be more complicated than simple time integration. The muscle forces required to rotate a joint depend on the load borne by the hand as well as the posture of the limb. Correspondingly complex and time varying properties would be required of a spinal network as the limb changed configuration and interacted with the environment. On the other hand, if the discharge rate of premotor neurons in RNm were to encode the time course of muscle activation rather than kinematic features of the movement, the computational requirements of the spinal cord would be reduced considerably. This view is supported by the many direct projections to motor neurons demonstrated by spike-triggered averaging from both M1 (Fetz and Cheney 1980) and RNm (Mewes and Cheney 1991; Sinkjaer et al. 1995), which seem unlikely to mediate a significant dynamical transformation.
. The ambiguity in interpretation of a signal expressing more than one parameter was addressed in another study of planar reaching movements. In this multiple regression study, M1 discharge was correlated first with movement direction, then target position, and finally movement distance (Fu et al. 1995). The authors made the suggestion that by such "temporal parcellation" of its varied kinematic features, the nature of the command signal at any given time could be sorted out by a downstream area. This would seem, however, a rather difficult task to relegate to the neural circuitry of the spinal cord.
, most M1 neurons projecting through the pyramids preferentially encoded forcelike variables rather than positionlike variables during the execution of simple wrist movements. Later work by Cheney and Fetz (1980) that considered only monosynaptically connected M1 neurons also found a strong correlation with force during wrist flexion/extension. It is not unreasonable that the correlation with force may have reflected muscle activation commands. Recent recordings comparing the timing and magnitude of M1 discharge with arm and hand muscle activity during movement in varied directions (Miller et al. 1996; Scott 1997) and intrinsic hand muscle activity during fractionated finger movements (Bennett and Lemon 1996) are consistent with this hypothesis.
; Miller et al. 1993). In the current study, analysis was restricted to strongly correlated unit/muscle pairs to provide the clearest picture of the dynamics of the rubrospinal pathway. Data were included from two different types of tasks: free-form movements and a single degree of freedom tracking task. We made comparisons across these tasks of the strength and duration of the bursts of RNm discharge and of the shape, width, and timing of cross-correlations calculated between RNm and electromyograms (EMG). Our results suggest that during neither task is there a significant dynamical transformation between RNm discharge and muscle activation. A model is presented that accounts for small task-related differences in correlation shape and width in terms of timing differences across the population of recorded neurons. These results are consistent with our earlier observations that RNm discharge directly encodes muscle activation (Miller and Houk 1995; Miller et al. 1993).
METHODS
Abstract
Introduction
Methods
Results
Discussion
References
. Alternatively, the monkeys operated any of six different manipulanda, each requiring movement about different joints to move a cursor trace between alternating target positions displayed on an oscilloscope (Gibson et al. 1985a; Miller and Houk 1995). Various devices required movements of the fingers, wrist, elbow, and shoulder as well as supination/pronation and a twisting movement of the hand and wrist. Each monkey was trained to perform all of the tasks and would switch readily among them.
.
. The RNm is recognized easily as the group of cells related to contralateral limb movements and located lateral and caudal to the eye-movement related cells of the oculomotor complex. During the last week of recordings, electrolytic lesions were placed at several recording sites. After all recordings were completed, the monkey was administered a lethal dose of pentobarbital sodium and perfused with physiological saline followed by 10% neutral buffered formalin. The brain was removed and blocked in the plane of an electrode left in the tissue. Forty-µm-thick frozen sections were cut in the frontal plane, and alternate sections were mounted and stained with luxol fast blue and cresyl violet. Electrode tracks and anatomic boundaries of the nucleus were identified, and their locations compared with observations made during the recordings. The limb containing the EMG electrodes was dissected, and the location of each recording electrode was confirmed.
. A detailed description of these algorithms as we have applied them has been published (Houk et al. 1987). A more general discussion of the application of these methods to physiological systems is also available (Marmarelis and Marmarelis 1978). To classify the shapes of the many cross-correlograms, we computed the cross-correlation between the experimental correlogram and each of four standard templates. A group of 20 neurons was selected from monkey AL that were particularly well correlated with at least one muscle or with velocity. Several of the more technical analyses comparing cross-correlations and more traditional methods were applied only to the files from this group of neurons.
RESULTS
Abstract
Introduction
Methods
Results
Discussion
References
View larger version (27K):
[in a new window]
FIG. 1.
Discharge of a single magnocellular red nucleus (RNm) neuron recorded during 2 different tasks. A: 3 repetitions of a prehension movement in which the monkey reached toward (R) and grasped (G) raisins presented in front of him. B and C: discharge rate and position recorded during a single degree-of-freedom tracking task requiring flexion/extension wrist movements. Bursts occurred for both directions of movement and were of short-duration and somewhat lower amplitude than for the prehension movements.
View larger version (31K):
[in a new window]
FIG. 6.
Segment of data recorded while the monkey reached for and grasped a raisin. There were numerous different patterns of muscle activity spanning the several seconds necessary to perform the movement. Activity of the lumbrical muscle bore close resemblance to that of the neuron.
. Measurements we made from a group of 20 strongly correlated neurons from monkey AL revealed a mean onset latency of 100 ms and a mean offset latency of 115 (discharge terminating after movement). This negative value was largely the result of several cases with quite large negative latencies. These cases, and perhaps others, appeared to be caused by cocontraction that persisted beyond the end of the movement (like that shown in Fig. 4), to which the RNm signal remained related.
View larger version (25K):
[in a new window]
FIG. 4.
Segment of data recorded during a tracking movement from the middle to the fully supinated position. Movement was preceded by an initial large burst of RNm activity that returned slowly to spontaneous level after ~1 s. Extensor digitorum communis (EDC) was well correlated with RNm in general, and in this trial in particular, its time course was remarkably similar to that of RNm. Muscle activity of both EDC and extensor carpi ulnaris was consistent with slowly declining cocontraction after the movement.
; which occurred during a trial in which the monkey moved initially in the wrong direction) firing rate increased directly with the velocity of joint rotation for both flexion and extension. There was no apparent relation between firing rate and the angular position of the device (Fig. 2B).
View larger version (20K):
[in a new window]
FIG. 2.
Relation between the RNm discharge and movement signals from Fig. 1. A, C, and D: strong relation demonstrated between discharge rate and the velocity of wrist flexion and extension movements. Large cross-correlation between RNm and the negative (flexion) components of the velocity signals (C) corresponds to the large negative slope of the relation shown in A. Weaker cross-correlation with positive (extension) movements (D) is reflected in the lower, positive slope in A. There was no significant correlation with position revealed by either scatter plot (B) or cross-correlation for this neuron.
max = 0.23, peak at 100 ms), consistent with the results of Fig. 2A.
View larger version (17K):
[in a new window]
FIG. 3.
Relation of the time of the peak RNm/Vel cross-correlation vs. the latency between the onset of discharge and movement for 20 files recorded from monkey AL. Correlations were calculated after having removed intertrial periods. Single anomalous data point ( ) was excluded from the calculation of the correlation coefficient.
1 s. Not only would the time of the offset of discharge be difficult to determine meaningfully, more importantly, its magnitude clearly did not encode movement velocity, which returned to zero in <200 ms.
View larger version (24K):
[in a new window]
FIG. 5.
Cross-correlations for data from the preceding figure. The large RNm/EDC cross-correlation reflects the similarity between RNm activity and the activation of EDC. RNm response also was well correlated with positive rectified velocity although the asymmetrical shape indicates that RNm frequently tended to be active beyond the end of the movement. Note the different scale of the vertical axis for the RNm autocorrelation (top left).
max = 0.17) because positive correlations in one direction were canceled by negative correlations in the other direction. Consequently, we full-wave rectified the velocity signal, making it bidirectional. The peak of this correlation reached 0.38 and occurred because the initial pulse of RNm discharge correlated well with the magnitude of the velocity of movement. The asymmetrical, slowly decreasing correlation at increasing negative lags was indicative of correlated RNm activity lagging velocity. This was caused by the prolonged RNm discharge that often persisted well beyond the end of each movement.
max = 0.49 at 75 ms), and the correlation with first dorsal interosseous was also quite strong (max = 0.42). Weaker correlations were obtained for several other muscles. The correlations in this figure exemplify several of the fundamental differences between the correlations typically found for the tracking and free-form tasks. The free-form correlations were typically larger and most were somewhat broader than those during tracking. A larger range of peak-correlation times (which in this example varied from 360 to 15 ms if one includes the weakly correlated muscles) was also characteristic of the free-form tasks. The greater breadth is at least in part due to the longer duration bursts of RNm activity. The relatively broad shoulders of the autocorrelation in Fig. 7, top left, compared with that of Fig. 5, reflect this greater burst duration.
View larger version (22K):
[in a new window]
FIG. 7.
Cross-correlation functions for the file from which the data in Fig. 6 were taken. Strongest correlation was with lumbrical, with a variety of smaller correlations resulting for other muscles involved in the task.
. An advantage of the cross-correlation is that it requires considerably less computer time to calculate. The major drawback is that, unlike the impulse response, the cross-correlation is affected by the spectrum of the input signal as well as the transformation between input and output signals. However, if the spectrum of the input signal (in this case RNm discharge rate) is approximately white, the two calculations are equivalent.
such that a brief input gives rise to relatively long-lasting output effects. The integrator shape shown in Fig. 8E was unusual for cross-correlations with EMG signals. As in these examples, the impulse responses were always considerably noisier than cross-correlations, often somewhat narrower, but otherwise of very similar shape.
View larger version (38K):
[in a new window]
FIG. 8.
Comparison of cross-correlation (thick lines) and impulse response functions (thin lines). Although the cross-correlations were generally much smoother, their shape was generally very similar to the envelope of the impulse response if occasionally somewhat broader.
View larger version (15K):
[in a new window]
FIG. 9.
Template matching comparison of statistically significant ( max 0.15) cross-correlation shapes. A: narrow pulse, broad pulse, pulse-step, and integrator type templates. B: distribution of template matches for tracking and free-form tasks. RNm/electromyographic (EMG) cross-correlations matched predominantly the broad pulse template (2) during both behaviors. There were very few pure integrator shapes (4).
0.25 that matched templates 1 or 2, the widths were approximately normally distributed, with means of 760 ± 380 ms and 970 ± 370 ms for the tracking and free-form, respectively. This difference resulted at least in part from similar differences in the autocorrelation width. However, the particular distribution between templates 1 and 2 is not of as much significance as the fact that the distribution of these templates for the two tasks was nearly identical and that matches with the pulse-like templates were much more frequent than with either of the templates with a step component.
0.25, which was well above statistical significance. We considered only those correlations that matched templates 1, 2, or 3, as the timing of the peak correlation would not have been meaningful for those correlations of marginal statistical significance or lacking a significant pulse component. Figure 10 shows distributions of these lags for monkeys BR and AL in A and B, respectively. The black bars indicate lags that occurred during the free-form tasks, while the hashed bars indicate data from the tracking tasks. The modes of the distributions were identical for the two monkeys, being ~50 ms for tracking and between 0 and 20 ms for free-form movements. Sixty-five percent of the lags fell between 100 and 200 ms during free-form movements and 80% during tracking.
View larger version (30K):
[in a new window]
FIG. 10.
Distribution of time to peak for RNm/EMG cross-correlations for strong ( max 0.25), phasic (templates 1, 2, or 3) cross-correlations. A positive lag indicates RNm leading EMG. A: monkey BR; B: monkey AL. Lags during free-form tasks were distributed more broadly and had a mode nearer 0 than during tracking.
; Mewes and Cheney 1994). Clearly, there exists a significant amount of well-correlated activity among different neurons during these tasks, separated by latencies on the order of 100 ms. If the activity in these different neurons were sufficiently well correlated and each contributed power to the EMG at slightly different lags, the shape and timing of both the cross-correlation and impulse response might be significantly affected.
View larger version (21K):
[in a new window]
FIG. 11.
Model of the coordinated control from multiple, correlated neurons. In all panels, the data were obtained from simulations using a real RNm signal as an input. A: output was formed within a spinal circuit of interneurons and motor neurons (IN/MN) by summing 1 identical input signals (RNi), each shifted by a random amount, together with an uncorrelated noise signal (N) of similar power and spectrum. Combined signal then was delayed by 75 ms. B and C: cross-correlation and impulse response functions for such a system with a single input. Peak correlation was <1.0 because of the added noise, and the width of the function reflects the input spectrum. However, the impulse response reveals that the model system contained no dynamics. Time of the peak of both functions resulted from the 75-ms delay. D and E: system with 5 correlated input neurons caused the cross-correlation to broaden, but individual peaks, centered on 75 ms were still noticeable. Impulse response clearly showed the 5 separate signals. F and G: cross-correlation for a 100-neuron system had a smooth, symmetrical shape very much like that of real data. Impulse response no longer had discernible individual peaks but rather a broad, noisy peak centered on 75 ms, also closely resembling that of the real data.
DISCUSSION
Abstract
Introduction
Methods
Results
Discussion
References
; Gibson et al. 1986b; Mewes and Cheney 1994) remain controversial issues. Our results are consistent with those earlier reports suggesting that although the predominant character of the RNm discharge is phasic, there is a small tonic component as well. The relative prominence of the phasic component is also true of M1 discharge during wrist movements although perhaps to a lesser extent (Cheney et al. 1988). One explanation for these findings is that RNm is specialized to produce relatively large muscle activation during limb movement, with a disproportionate contribution to maintenance of limb posture from another area of the cortex or brain stem.
. These and the wrist extensor muscles also have been shown by spike-triggered averaging to be the most frequent targets of monosynaptic RNm projections (Mewes and Cheney 1991; Sinkjaer et al. 1995). During reaching and grasping movements in particular, these muscles tend to be activated in short bursts of activity. If a disproportionate number of RNm neurons control these muscles, it would not be at all surprising that RNm discharge also should have a strongly phasic character.
have shown that for normal speed, whole arm movements, the velocity-related forces generated between limb segments are of the same order of magnitude as tonic, gravity-related forces. During fast movements, these dynamic intersegmental forces dominate those due to gravity. During our free-form tasks, some part of the limb was moving throughout each task. The greatest activity nearly always occurred in bursts, clearly associated with a phase in which the muscle was acting to move the associated limb segment. The most proximal limb muscles, for example, anterior deltoid, biceps, and triceps, did reveal obvious sustained activity during the period in which the limb was extended. This amplitude, however, was typically smaller than the phasic activity of these same muscles during the proximal limb movement.
. Our results, which address the voluntary control of movement, in no way contradict these observations. The descending voluntary movement commands from the RNm undoubtedly are combined with input from other descending systems as well as cyclic and reflexive input generated at the spinal level. At issue is simply the coordinate system in which these convergent signals are encoded.
; Mewes and Cheney 1991; Sinkjaer et al. 1995) is an order of magnitude shorter than the 50- to 100-ms delay between onset of discharge and whole-muscle EMG (Gibson et al. 1985b; Mewes and Cheney 1994; Thach 1975). It has been suggested that the additional time may be necessary to bring interneurons and motor neurons to threshold at the onset of movement. This process may account for some of the excess of phasic neuronal discharge over that of muscles at the onset of a tracking movement (Fetz et al. 1989). Gibson and his colleagues (1985b) noted that the otherwise linear relation between discharge rate and the velocity of joint rotation included a significant offset in discharge rate above zero velocity. This nonlinearity in the rate/velocity curve may represent the same phenomenon.
. Others have noted that the range of onset latencies also become considerably greater for more complex movements or movements to which a given neuron is less well related (Cheney and Fetz 1980; Gibson et al. 1985a; Thach 1975). The unusually long and negative latencies may result from neurons that actually control some aspect of the movement other than its onset. It is likely that the occurrence of the long and negative cross-correlation lags has a similar explanation. For example, a neuron that is functionally related to some unrecorded muscle also will be correlated with synergist muscles active in other phases of the behavior. During the tracking task, muscles tended to be activated nearly simultaneously, contributing a relatively restricted range of unit/muscle lag times. There was a considerably broader range of onset and offset times for muscles during the free-form tasks. Most notably, the intrinsic hand muscles were activated late in the behavior as the monkey was grasping the piece of fruit. As a group, these muscles contributed a significant number of long, positive lags that did not occur during the tracking tasks and skewed the mean lag in the positive direction during the free-form tasks.
either positive or negative. Furthermore, the fact that the shape of the correlations did not depend greatly on the time of the peak argues for a single mechanism causing both the positive and negative lag correlations.
. The model is capable of learning to produce movement commands in an intrinsic, muscle-based coordinate system, much like that predicted by these results. Such a system would be capable of learning to modify its movement commands to reflect the widely changing mechanical properties of the limb during its interactions with the environment. The result would be the relatively simple dynamics that we have demonstrated between RNm discharge and muscle activity.
; Cheney and Fetz 1980; Evarts 1969; Hepp-Reymond et al. 1978; Kalaska et al. 1989; Miller et al. 1996; Scott 1997). Additional efforts must be made to elaborate specific force or muscle space models for M1. Elucidation of such a simple encoding scheme, common to both the motor cortex and red nucleus would inevitably lead to a better understanding of the limb motor systems in general.
| |
ACKNOWLEDGEMENTS |
|---|
The authors acknowledge the assistance of T. Andersen, who participated in many of the experiments, and J. Houk, who offered helpful suggestions for improving the manuscript.
This work was supported in part by National Institute of Mental Health Grant 5 P50 MH-48185. T. Sinkjaer was supported in part by a grant from the Danish National Research Foundation.
| |
FOOTNOTES |
|---|
Address for reprint requests: L. E. Miller, Physiology Dept., Northwestern University Medical School, 303 E. Chicago Ave., Chicago, IL 60611.
Received 17 April 1997; accepted in final form 19 February 1998.
| |
REFERENCES |
|---|
|
Abstract
Introduction
Methods
Results
Discussion
References
|
|---|
This article has been cited by other articles:
|
|
 |
|
  K. Yamamoto, M. Kawato, S. Kotosaka, and S. Kitazawa Encoding of Movement Dynamics by Purkinje Cell Simple Spike Activity During Fast Arm Movements Under Resistive and Assistive Force Fields J Neurophysiol, February 1, 2007; 97(2): 1588 - 1599. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. M. Morrow, L. R. Jordan, and L. E. Miller Direct Comparison of the Task-Dependent Discharge of M1 in Hand Space and Muscle Space J Neurophysiol, February 1, 2007; 97(2): 1786 - 1798. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. R. Townsend, L. Paninski, and R. N. Lemon Linear Encoding of Muscle Activity in Primary Motor Cortex and Cerebellum J Neurophysiol, November 1, 2006; 96(5): 2578 - 2592. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Shah, A. H. Fagg, and A. G. Barto Cortical Involvement in the Recruitment of Wrist Muscles J Neurophysiol, June 1, 2004; 91(6): 2445 - 2456. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. M. Morrow and L. E. Miller Prediction of Muscle Activity by Populations of Sequentially Recorded Primary Motor Cortex Neurons J Neurophysiol, April 1, 2003; 89(4): 2279 - 2288. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Lavoie and T. Drew Discharge Characteristics of Neurons in the Red Nucleus During Voluntary Gait Modifications: A Comparison with the Motor Cortex J Neurophysiol, October 1, 2002; 88(4): 1791 - 1814. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
P. L. E. Van Kan and M. L. McCurdy Role of Primate Magnocellular Red Nucleus Neurons in Controlling Hand Preshaping During Reaching to Grasp J Neurophysiol, April 1, 2001; 85(4): 1461 - 1478. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Visit Other APS Journals Online |
