- Email: [email protected]
Peripheral regulation of stiffness during arm movements by coactivation of the antagonist muscles
Brain Research, 473 (1988) 134 14(~ Elsevier
134 BRE 23130
Short Communications
Peripheral regulation of stiffness during arm movements by coactivation of the antagonist muscles Roger W. Simmons 1 and Charles Richardson 2 IMotor Performance Laboratory, San Diego State University, San Diego, CA (U.S.A.) and 2Naval Ocean Systems Center, San Diego, CA (U.S.A.)
(Accepted 21 June 1988) Key words: Mechanical impedance; Arm movement; Electromyographic pattern
Two experiments investigated whether unexpected and differential loading of a rapid, unsighted arm movement resulted in the central nervous system (CNS) regulating limb stiffness by modifying the associated neuromuscular activity. In Experiment 1, subjects completed multiple, spring-loaded training trials until a prespecified criterion of learning was attained. On selected trials, the spring load was unexpectedly replaced by an inertial load. Results indicated that to maintain positional accuracy during this inertial load trial, limb stiffness was increased by coactivating the antagonist muscles, i.e. b~ changing the associated neuromuscular activity from a predominantly triphasic pattern to one of coactivation. In Experiment 2, the sequence of loading was reversed producing a change in the required limb stiffness from a relatively high to low level. This change was observed as a pattern of coactivation being replaced by a triphasic activity pattern. These results support the notion that limb stiffness is regulated primarily through modification of the neuromuscular activity pattern prior to movement termination. It was also demonstrated that the size of the unexpected load did not affect the basic activation pattern selected by the CNS. It is proposed that the signal which triggers the CNS to regulate limb stiffness is based on peripheral information generated as a result of agonist activity occurring during the first part of the movement. O n e avenue of research currently pursued by individuals interested in m o t o r control, centers on identifying the variable(s) controlled by the central nervous system (CNS) during the generation of a m o t o r response 15. To date, muscle length I4, velocity 1°, acceleration 16, force 2, stiffness 9 and various combinations of the above 7 have each been p r o p o s e d as centrally controlled variables. With specific reference to the variable of stiffness, Hogan s has described a theory in which m o t o r control is m a i n t a i n e d by adaptively coactivating the antagonist muscles. Accordingly, the CNS 'presets' the activation patterns of the muscles, in advance of the response, for the purpose of generating the appropriate level of stiffness necessary to control the movement. Subsequent to each response, the appropriateness of the actual stiffness c o m p o n e n t , and the accuracy of the response, are evaluated in terms of the desired stiffness and response outcome through the use of intrinsic feedback and knowledge of results
information. If the actual level of stiffness is determined to be i n a p p r o p r i a t e , the CNS resets the muscle activation pattern. This adaptive response is used to p r o d u c e a new and more a p p r o p r i a t e stiffness component for controlling the next movement. A p a t t e r n of coactivation is particularly suited to any type of response requiring relatively high levels of stiffness such as stabilizing the body or body segments against the force of gravity s. Coactivation is also a viable pattern for generating the a p p r o p r i a t e level of stiffness required to grip an object or to move a l o a d e d e x p e r i m e n t a l apparatus, although these represent m o r e complicated examples of stiffness reguiation 8. In these latter types of response, a mechanical interaction occurs between the hand and object or apparatus that provides information about the dynamical properties of the complete response system (i.e. arm, hand, object or apparatus). Associated with the object or apparatus is a stiffness comp o n e n t that theoretically ranges from zero to a level
Correspondence." C. Richardson. Present address: P.O. Box 544, Rancho, Santa Fe, CA 92067, U.S.A.
0006-8993/88/$03.50 © 1988 Elsevier Science Publishers B.V. (Biomedical Division)
135 that prevents the object or apparatus from being moved. The difference between this apparatus stiffness and the total stiffness required to control the movement must be generated by the muscles of the limb. For example, if the object or apparatus provides little or no inherent stiffness, the stiffness required to control the response must be entirely provided by the limb musculature. On the other hand, if a relatively high stiffness component is provided by the object or apparatus, the need for the limb muscles to provide the necessary stiffness is considerably reduced. According to the foregoing notion, coactivation would be appropriate for movements requiring either high or low limb stiffness for effective motor control. However, the use of a coactivation pattern to generate relatively low levels of limb stiffness incurs a high metabolic cost 8. To offset this physiological inefficiency, coactivation is changed to a triphasic activation pattern capable of producing the same degree of motor control but with less energy expenditure. A triphasic activation pattern is characterized by two discrete pulses of agonist activity separated in time by a single antagonist burst. A key feature of Hogan's theory of motor control is that in order to generate the required level of stiffness, the nature of the mechanical interaction between the limb and the object or apparatus must be known to the subject prior to response execution. It is not clear what happens to the control system when the response is unexpectedly perturbed by a load causing the nature of the mechanical interaction to completely change. Many studies have used an experimental paradigm in which the magnitude of inertial loads has been unexpectedly increased or decreased 3'4A3, but altering load size does not change the mechanical interaction between limb and apparatus. A change in the nature of the interaction could be induced if the type (as opposed to the size) of load were varied. To our knowledge, only one study has used different types of loads to perturb a response but the investigators did not record the electromyographic (EMG) activity underlying each movement 11. Therefore, the purpose of the present experiments was to investigate the regulation of stiffness during the execution of a rapid arm movement unexpectedly perturbed by different types of loads. In the first ex-
periment, an elastic spring load was applied to the apparatus. Since this load possesses a high stiffness component, the requirement for additional limb stiffness to produce the desired movement control is greatly reduced, resulting in the adoption of a triphasic activation pattern. On selected trials the familiar spring load was unexpectedly replaced by an inertial load possessing no stiffness. Thus, it was predicted that in order to generate a new and higher level of limb stiffness necessary to stabilize the inertial load at the target, the triphasic activation pattern would be changed to coactivation. Twelve right-hand dominant male university students, placed their right forearm on a wooden lever (48.5 cm long) attached to a base board by a nearfrictionless metal bearing (3.80 cm radius). The olecranon process was positioned directly above the axis of rotation. The hand was placed flat over a wooden block situated at the distal end of the lever. The angle of the right forearm at the movement start position relative to the frontal plane of the body was 0.8727 rad. The center of a spatial target area was marked on the base board at 1.674 rad with upper and lower boundaries of + 0.055 rad. This arrangement defined a movement arc of 0.8013 rad. A mechanical load was applied to the lever arm in one of two ways. An inertial load consisted of a small lead block placed in an aluminum cradle mounted at the distal end of the lever. Three inertial loads were constructed to weigh 1.1, 1.5 and 2.0 kg. An elastic spring load was applied via a thin metal wire attached to the lever 25.5 cm distal from the center of rotation. The wire passed around a freely rotating pulley (4.5 cm radius) positioned 20.3 cm to the right of the lever and attached to a linear slide mechanism positioned directly in line with the wire. The 35-cm-long spring load was attached to a small metal hook fixed to the slide and a second hook secured the spring to the apparatus base board. A Teflon cradle placed between the hooks supported the spring load. Movement of the lever through 0.8013 rad to the target center produced a linear spring extension of 18 cm. Three springs were constructed of wire and designated according to the force required to stabilize the spring at 18 cm. This calibration resulted in spring loads of 1.1, 1.5 and 2.0 kg with stiffness values of 0.06, 0.08 and 0.11 kg/cm, respectively. A 25-cm linear potentiometer was attached to the
136 slide and provided an electrical analog of the displacement of the slide (i.e. a linear measure of the rotational displacement of the lever arm). The electrical analog was recorded on a Honeywell Visicorder at a paper speed of 63.5 cm/s. The output of the potentiometer was calibrated to the center and upper and lower boundaries of the target area prior to each testing session. E M G activity was monitored by two pairs of surface electrodes (Beckman) positioned parallel to the long axis of the bicep (agonist) and two electrode pairs over the tricep (antagonist) muscles. The interelectrode distance was approx. 3.0 cm, and interelectrode resistance less than 10 kQ for each electrode pair. A ground electrode was placed over the right acromion process. Four channels of raw E M G signals were processed using 4 Beckman couplers (Type 9852A) in combination with a Beckman R-411 6-channel recorder. Amplifiers were calibrated prior to each testing session, and the high frequency response set at maximum. Analog signals were displayed using curvilinear pens and a paper speed of 100 mm/s. Blindfolded subjects were required to complete a series of training trials until the response was 'learned'. This was achieved when the subject completed 5 consecutive trials with each trial terminated in the target area in less than 200 ms. To facilitate acquisition of the skill knowledge of results about the magnitude and direction of the response error (e.g. +1.5 U) and the speed of the movement was provided after each trial. All training trials were completed with a 1.1 kg control spring load attached to the slide. After the 5th successful training trial, one more movement was completed and designated a test load trial. Time-displacement and E M G data were recorded during each test load trial. This sequence of training trials followed by a single test load trial was repeated 8 times by each subject. For each test load trial, either the 1.1 kg control spring load was removed and then reapplied to the slide or was replaced by a 1.1, 1.5 or 2.0 kg inertial test load. The procedures for interchanging the spring and inertial loads were identical for each test load trial. The spring load was randomly applied on 4 of the test load trials, and one of the inertial loads applied on the other 4 test load trials. Four subjects were ran-
domly assigned to one of 3 independent groups. All groups trained with the spring load, but differed in the size of the inertial load. Groups 1, 2 and 3 represent inertial test loadings of 1.1, 1.5 and 2.0 kg, respectively. Each E M G record was evaluated by two raters on two separate occasions (at least 24 h apart), and classified as being either coactivation or triphasic in nature. The criteria used for classification were as follows. A triphasic pattern was characterized by two discrete bursts of agonist activity separated by a silent period, during which a distinct pulse of antagonist activity was observed. A typical triphasic pattern observed during spring load conditions is presented in Fig. 1A. When a single burst of agonist activity occurred simultaneously with a sustained and relatively constant level of innervation of the antagonist muscle, it was considered to be characteristic of a coactivation pattern. An example of this E M G pattern observed when the movement was unexpectedly loaded by an inertial load is illustrated in Fig. lB. Inter-rater reliability for classifying'EMG patterns during spring load trials and inertial load trials was Ca = 0.609 and 0.483, respectively. These reliability coefficients were transformed into ~ statistics 5 and found to be significant, ~ (1) = 17.8 and 11.2, P < 0.01. For spring load trials, 69% of the E M G patterns were classified as triphasic patterns, whereas on inertial load trials 94.0% were classified as coactivation patterns. There were no differences in frequency or type of E M G patterns as a function of load (i.e. groups 1,2, and 3). The data of Experiment 1 support the predicted relationship between the type of external load force and the associated E M G activity. Practising the movement with an elastic spring load attached to the apparatus presumably biased the CNS to preset the musculature to generate a triphasic activation pattern. However, when the response was unexpectedly perturbed by an inertial load, the results indicated that the CNS reset the neuromuscular activation pattern prior to movement termination to produce an optimal control pattern of coactivation. A second experiment was conducted to determine if the principle of modifying the neuromuscular innervation pattern observed in the first experiment occurs when the sequence of loading is reversed (i.e.
137 inertial
to spring load). The procedures
periment
2 were identical
used in Ex-
to those adopted in the first
experiment with one exception. Twelve subjects not used in Experiment 1, completed the training trials with a 1.1 kg inertial load attached to the apparatus. On test load trials, the inertial load was detached and then reattached to the apparatus or was replaced by a 1.1,1.5or2.0kgspringload.
Each EMG record for the test load trials was evaluated using procedures
and waveform
criteria identi-
cal to those used in Experiment 1. Typical coactivation and triphasic patterns produced by subjects during the second experiment are presented in Fig. 2A,B. Inter-rater reliability for classifying EMG patterns for the inertial 0.342
and
0.295,
and spring load trials was q = respectively.
These
coefficients
0
A
AVG. DISP
in mVOLTS
AVG. DISP.
0.4375,
in mVOLTS
0.4375,
0.0625 t
EMG
EMG
UNITS
I
0.51
0.375 0.25
0.75 ‘I-
1 I
illl
-0.25
I
0.S
u
0 -0.125
UNITS
I
II
0.125
/
0.2s 0 -0.25 -0.5
-0.5 r -0.375
-Dd 0 EMG
UNITS
EMG
I
0.375
I
1
400
500
-0.25L
1.51
1 100
1 200
I 300
I 400
500
I 100
1 200
I 300
1 400
I 500
UNITS
-1.5 -2-2.5 D
L
1
100
200 TIME
, 300 (mSEC
1
0
TIME
(mSEC)
Fig. 1. Average time-displacement and EMG activation patterns associated with rapid arm movements perturbed by a control spring load (A) or a test inertial load (B). The top trace, displacement; middle trace, agonist EMG record; lower trace, antagonist EMG record. All 3 records are the average of 4 single records for a subject in Group 2. The spring load was 1.0 kg and the test inertial load was 2.0 kg. Average displacement is marked in mV with 1.0 mV representing 2.13 rad. The two horizontal lines in the displacement graph represent the target area. The upper and lower traces are time-locked to the onset of agonist activity and the time base is expressed as beginning 50 ms prior to this point. The difference between the records in A and B is most noticeable in the activity of the antagonist. In A a single discrete antagonist burst is consistent with a triphasic pattern. In B antagonist activity is sustained throughout the last half of the movement and indicates a coactivation pattern.
138 were statistically significant, Z: (1) = 5.61 and 4.12, P < 0.05. For inertial test load trials, 83% of the E M G patterns were observed to be coactivation and 79% of the spring test load trials were classified as triphasic. The limb-apparatus system has a total stiffness component equal to the sum of the stiffness inherent in both the limb and the apparatus. Thus, to produce an efficient and accurate m o v e m e n t to a target, the limb stiffness must be scaled to the mechanical stiffness of the apparatus. An error in scaling will result in either an underdamped or overdamped system,
which in turn will produce inaccurate positional control and significant alteration in m o v e m e n t velocity and acceleration. In Experiment 1, the relatively large stiffness component provided by the spring load during the testload trials reduced the need for limb-generated stiffness and motor control was maintained using a more energy-efficient triphasic pattern. H o w e v e r , during unexpected inertial loading, this relatively low level of limb stiffness generated in anticipation of the spring load represented an underdamped response system, which if left unchanged would have resulted
AVG. DISP. in m V O L T S
AVG. DISP. in reVOLTS
0.4375
0,437~.
f
0.375
0.375 0.3125
0.3125 0.25
0.25
0.1875
0.1875
0.125
0.125
0.0625
0.0625 I
100
0
200
300
0
400
500 EM(
UNITS
EM,
0
0,75
(175
0.5
0.5
0.25
0.25 C
0.25
-0.25
-0.5
-0.5
0 EMG
-0.75
I
1
I
t
100
200
300
400
500
0.5
1
0.25
0.5
e
O
-o.25
-0.5
-o.5
0
0
-o.75
I
1
I
100
200 TIME
300 (rnSEC)
400
500
300
400
500
lOO
200
3oo
400
500
I
I
200
300
400
UNITS
EM(
NITS
1.5
200
UNITS
C
-0.7 ~_
100
I
o
100
500
TIME (mSEC)
Fig. 2. Average time-displacement and EMG activation patterns associated with rapid arm movements perturbed by a control inertial load (A) and a test spring load (B). The records are averages for a subject in Group 1. Inertial load was 1.0 kg and the springload 1.0 kg. Tracings of markings for the graphs are the same as in Fig. 1. Under inertial load conditions (A), the antagonist indicates coactivation while during unexpected loading with a spring load the antagonist conforms to a triphasic activation pattern,
139 in an overestimation of the target. To achieve the desired positional accuracy, limb stiffness needed to be increased which, as predicted, was accomplished by the CNS resetting the neuromuscular activation pattern from a triphasic pattern to one of coactivation. In Experiment 2, the switch from an inertial to spring load required a decrease in limb stiffness. Failure to effect this change would have resulted in a severely overdamped response causing degradation of movement efficiency and accuracy. In this case, modification of limb stiffness was observed as a switch from a coactivation pattern to a triphasic pattern. Therefore, the results of Experiment 2 fully complemented the findings of the first experiment, and collectively the two experiments demonstrated that regardless of the sequence of loading, the CNS can modify the muscular innervation patterns prior to completion of a response. Another significant feature of the data was that increases in the size of the test load did not affect modification of the neuromuscular activation pattern. It has been previously demonstrated that increased loading does produce different levels of neuromuscular innervation 6, but according to the present results, amplitude modulation does not alter the overall configuration of the E M G waveform specified by the CNS. An interesting aspect to the present results concerns the nature of the signal that triggers the CNS to change the activation pattern from one type to another. A common explanation of rapid error correction during visually guided responses is provided by the concept of efference copy 17 or corollary discharge 12. Accordingly, error detection-error correction is thought to occur at a central level through the use of an 'internal' feedback loop. Response-produced feedback need not be included as part of this internal loop meaning that errors can be detected and corrected prior to movement initiation. The notion of modifying activation patterns through a mechanism such as efference copy or corollary discharge is not applicable to the unsighted and unexpectedly loaded movements used in the present experiments. Based on the assu/nption that the conditions of the training trials are constant, and in the absence of visual input to indicate that conditions have changed, the subject will generate a set of efferent commands founded on the expectation that the
mechanical interaction between the limb and load will be the same on the upcoming trial as it was for past training trials. Furthermore, the same set of efferent commands will be discharged to the musculature unless a difference in load is detected. However, detection of any differences in the stiffness properties of each load is only possible after movement initiation, which eliminates the use of a central system correcting errors prior to movement initiation. In the absence of prior information and visual input it seems reasonable to assume that changes in the mechanical interaction between limb and apparatus are detected by the proprioceptive system. This leads to the possibility that peripheral input drives the central structures to regulate limb stiffness in accordance with any detected alteration in external load conditions. The limitation with proposing a peripheralcentral interaction is to explain how such a feedback loop can work during rapid movements (i.e. movements completed in <200 ms). During these fast responses the transmission delays associated with information traversing the transcortical feedback loop often exceed the total movement time, so that no corrections were made to the response observed prior to movement termination. In Experiment 1, where the unexpected application of the inertial load increased movement time to 307.8 ms, sufficient time was available for this loop to be completed and correct modification of limb stiffness to be made. In Experiment 2, however, the unexpected imposition of the spring load resulted in movement time decreasing to 158.6 ms, which placed severe temporal limitations on the time available to complete the feedback loop. Since the regulation of limb stiffness cannot occur prior to movement initiation through a central mechanism such as efference copy, but must be completed fast enough to allow sufficient time for the new efferent commands to be enacted prior to movement termination, the peripherally generated signal to switch activation patterns must occur during a narrow window of time corresponding to the earliest part of the movement. A follow-up examination of the time of onset and time of maximal innervation for the agonist and antagonist muscles revealed that although innervation of the antagonist occurred 15 ms prior to movement initiation, the agonist was fully involved during the early part of the movement. The agonist was acti-
14() vated 83.8 ms prior to movement initiation and was maximally innervated at the point of m o v e m e n t onset. This involvement can be seen in Fig. 2, which also indicates that the major contribution of the an-
used visually guided movements, which as previously noted, represent a different class of responses. This leaves the contribution of the agonist pulse to the control of rapid, unsighted, isotonic movements with
tagonist does not occur until later in the response.
variable limb stiffness requirements virtually unknown.
These data and observations suggest that peripheral information generated primarily as a result of agonist activity during the first part of the m o v e m e n t
In summary, data are presented which demonstrate that limb stiffness associated with a rapid un-
is responsible for signaling the CNS to reset limb stiffness when a new load is encountered. This is not an
sighted response can be regulated prior to movement termination. Regulation was imposed through the
entirely new idea, having been previously proposed by Allum j. He suggested that during isometric con-
CNS resetting current neuromuscular activation to a pattern more conductive to producing the required
tractions against preexisting force levels, the agonist
limb stiffness. Furthermore, the signal to reset limb stiffness when an unexpected load type was encoun-
is used as a medium latency E M G response which provides a pulse-test signal about external loading conditions to the CNS. Based on the outcome of the pulse test, the muscles are activated to cater to the external load conditions. Other investigators have
tered appears to be composed of peripherally generated sensory information resulting from a high level of agonist activity occurring during the early part of the movement.
also studied the role of the agonist burst, but have
1 Allum, J.H.J., Responses to load disturbances in human shoulder muscles: the hypothesis that one component is a pulse test information signal, Exp. Brain Res., 22 (1975) 307-326. 2 Bawa, P.N.S. and Dickinson, J., Force as'the controlling muscle variable in limb movement, Behav. Brain Sci., 5 (1982) 543-544. 3 Bizzi, E., Dev, P., Morasso, P. and Polit, A., Effects of load disturbance during centrally initiated movements, J. Neurophysiol., 41 (1978)542-566. 4 Brown, S.H.C. and Cooke, J.D., Responses to force perturbations preceding voluntary human arm movements, Brain Research, 22 (1981) 350-355. 5 Guildford, J.P., Psychometric Methods, McGraw Hill, New York, 1954. 6 Hannaford, B., Lakshminarayanan, V., Stark, L. and Nam, M., Electromyographic evidence of neurological controller signals with viscous load, J. Mot. Behav., 16 (1984) 255-274. 7 Hoffer, J.A., Central control and reflex regulation of mcchanical impedance: the basis for a unified motor-control scheme, Behav. Brain Sci., 5 (1982) 548-549. 8 Hogan, N., Adaptive control of mechanical impedance by coactivation of antagonist muscles, IEEE Trans. Eng., 29 (1984) 681-690. 9 Houk, J.C., Regulation of stiffness by skeletomotor reflexes, Annu. Rev. Physiol., 41 (1979)99-114. 10 Lestienne, F., Effects of inertial load and velocity on the
braking process of voluntary limb movements, Exp. Brain Res., 35 (1979) 407-418. 11 Schmidt, R.A. and McGown, C.M., Terminal accuracy of unexpectedly loaded rapid movements: evidence for a mass-spring mechanism in programming, J. Mot. Behav., 12 (1980) 149-162. 12 Sperry, R.W., Neural basis of the spontaneous optokinetic response produced by visual inversion, J. Comp. Physiol. Psychol., 43 (1950) 482-489. 13 Simmons, R.W. and Richardson, C., Maintenance of equilibrium point control during an unexpectedly loaded rapid limb movement, Brain Research, 302 (1984) 239-244. 14 Stark, L., Neurological ballistic movements: sampled data or intermittent open-loop control, Behav, Brain Sci., 5 (1982) 564-566. 15 Stein, R.B., What muscle variable(s) does the nervous system control in limb movements? Behav. Brain Sci.. 5 (1982) 535-577. 16 Terzuolo, C.A., Soechting, J.F. and Palminteri, R., Studies on the control of some simple motor tasks. III. Comparison of the EMG pattern during ballistically initiated movements in man and squirrel monkey, Brain Research, 62 (1973) 242-246. 17 Van Hoist, E. and Minelstaedt, H., The reafference principle. Interaction between the central nervous system and the periphery, 1950. In Selected Papers of Erich yon Holst, The Behavioural Physiology of Animals and Man, Methuen, London, 1973.
134 BRE 23130
Short Communications
Peripheral regulation of stiffness during arm movements by coactivation of the antagonist muscles Roger W. Simmons 1 and Charles Richardson 2 IMotor Performance Laboratory, San Diego State University, San Diego, CA (U.S.A.) and 2Naval Ocean Systems Center, San Diego, CA (U.S.A.)
(Accepted 21 June 1988) Key words: Mechanical impedance; Arm movement; Electromyographic pattern
Two experiments investigated whether unexpected and differential loading of a rapid, unsighted arm movement resulted in the central nervous system (CNS) regulating limb stiffness by modifying the associated neuromuscular activity. In Experiment 1, subjects completed multiple, spring-loaded training trials until a prespecified criterion of learning was attained. On selected trials, the spring load was unexpectedly replaced by an inertial load. Results indicated that to maintain positional accuracy during this inertial load trial, limb stiffness was increased by coactivating the antagonist muscles, i.e. b~ changing the associated neuromuscular activity from a predominantly triphasic pattern to one of coactivation. In Experiment 2, the sequence of loading was reversed producing a change in the required limb stiffness from a relatively high to low level. This change was observed as a pattern of coactivation being replaced by a triphasic activity pattern. These results support the notion that limb stiffness is regulated primarily through modification of the neuromuscular activity pattern prior to movement termination. It was also demonstrated that the size of the unexpected load did not affect the basic activation pattern selected by the CNS. It is proposed that the signal which triggers the CNS to regulate limb stiffness is based on peripheral information generated as a result of agonist activity occurring during the first part of the movement. O n e avenue of research currently pursued by individuals interested in m o t o r control, centers on identifying the variable(s) controlled by the central nervous system (CNS) during the generation of a m o t o r response 15. To date, muscle length I4, velocity 1°, acceleration 16, force 2, stiffness 9 and various combinations of the above 7 have each been p r o p o s e d as centrally controlled variables. With specific reference to the variable of stiffness, Hogan s has described a theory in which m o t o r control is m a i n t a i n e d by adaptively coactivating the antagonist muscles. Accordingly, the CNS 'presets' the activation patterns of the muscles, in advance of the response, for the purpose of generating the appropriate level of stiffness necessary to control the movement. Subsequent to each response, the appropriateness of the actual stiffness c o m p o n e n t , and the accuracy of the response, are evaluated in terms of the desired stiffness and response outcome through the use of intrinsic feedback and knowledge of results
information. If the actual level of stiffness is determined to be i n a p p r o p r i a t e , the CNS resets the muscle activation pattern. This adaptive response is used to p r o d u c e a new and more a p p r o p r i a t e stiffness component for controlling the next movement. A p a t t e r n of coactivation is particularly suited to any type of response requiring relatively high levels of stiffness such as stabilizing the body or body segments against the force of gravity s. Coactivation is also a viable pattern for generating the a p p r o p r i a t e level of stiffness required to grip an object or to move a l o a d e d e x p e r i m e n t a l apparatus, although these represent m o r e complicated examples of stiffness reguiation 8. In these latter types of response, a mechanical interaction occurs between the hand and object or apparatus that provides information about the dynamical properties of the complete response system (i.e. arm, hand, object or apparatus). Associated with the object or apparatus is a stiffness comp o n e n t that theoretically ranges from zero to a level
Correspondence." C. Richardson. Present address: P.O. Box 544, Rancho, Santa Fe, CA 92067, U.S.A.
0006-8993/88/$03.50 © 1988 Elsevier Science Publishers B.V. (Biomedical Division)
135 that prevents the object or apparatus from being moved. The difference between this apparatus stiffness and the total stiffness required to control the movement must be generated by the muscles of the limb. For example, if the object or apparatus provides little or no inherent stiffness, the stiffness required to control the response must be entirely provided by the limb musculature. On the other hand, if a relatively high stiffness component is provided by the object or apparatus, the need for the limb muscles to provide the necessary stiffness is considerably reduced. According to the foregoing notion, coactivation would be appropriate for movements requiring either high or low limb stiffness for effective motor control. However, the use of a coactivation pattern to generate relatively low levels of limb stiffness incurs a high metabolic cost 8. To offset this physiological inefficiency, coactivation is changed to a triphasic activation pattern capable of producing the same degree of motor control but with less energy expenditure. A triphasic activation pattern is characterized by two discrete pulses of agonist activity separated in time by a single antagonist burst. A key feature of Hogan's theory of motor control is that in order to generate the required level of stiffness, the nature of the mechanical interaction between the limb and the object or apparatus must be known to the subject prior to response execution. It is not clear what happens to the control system when the response is unexpectedly perturbed by a load causing the nature of the mechanical interaction to completely change. Many studies have used an experimental paradigm in which the magnitude of inertial loads has been unexpectedly increased or decreased 3'4A3, but altering load size does not change the mechanical interaction between limb and apparatus. A change in the nature of the interaction could be induced if the type (as opposed to the size) of load were varied. To our knowledge, only one study has used different types of loads to perturb a response but the investigators did not record the electromyographic (EMG) activity underlying each movement 11. Therefore, the purpose of the present experiments was to investigate the regulation of stiffness during the execution of a rapid arm movement unexpectedly perturbed by different types of loads. In the first ex-
periment, an elastic spring load was applied to the apparatus. Since this load possesses a high stiffness component, the requirement for additional limb stiffness to produce the desired movement control is greatly reduced, resulting in the adoption of a triphasic activation pattern. On selected trials the familiar spring load was unexpectedly replaced by an inertial load possessing no stiffness. Thus, it was predicted that in order to generate a new and higher level of limb stiffness necessary to stabilize the inertial load at the target, the triphasic activation pattern would be changed to coactivation. Twelve right-hand dominant male university students, placed their right forearm on a wooden lever (48.5 cm long) attached to a base board by a nearfrictionless metal bearing (3.80 cm radius). The olecranon process was positioned directly above the axis of rotation. The hand was placed flat over a wooden block situated at the distal end of the lever. The angle of the right forearm at the movement start position relative to the frontal plane of the body was 0.8727 rad. The center of a spatial target area was marked on the base board at 1.674 rad with upper and lower boundaries of + 0.055 rad. This arrangement defined a movement arc of 0.8013 rad. A mechanical load was applied to the lever arm in one of two ways. An inertial load consisted of a small lead block placed in an aluminum cradle mounted at the distal end of the lever. Three inertial loads were constructed to weigh 1.1, 1.5 and 2.0 kg. An elastic spring load was applied via a thin metal wire attached to the lever 25.5 cm distal from the center of rotation. The wire passed around a freely rotating pulley (4.5 cm radius) positioned 20.3 cm to the right of the lever and attached to a linear slide mechanism positioned directly in line with the wire. The 35-cm-long spring load was attached to a small metal hook fixed to the slide and a second hook secured the spring to the apparatus base board. A Teflon cradle placed between the hooks supported the spring load. Movement of the lever through 0.8013 rad to the target center produced a linear spring extension of 18 cm. Three springs were constructed of wire and designated according to the force required to stabilize the spring at 18 cm. This calibration resulted in spring loads of 1.1, 1.5 and 2.0 kg with stiffness values of 0.06, 0.08 and 0.11 kg/cm, respectively. A 25-cm linear potentiometer was attached to the
136 slide and provided an electrical analog of the displacement of the slide (i.e. a linear measure of the rotational displacement of the lever arm). The electrical analog was recorded on a Honeywell Visicorder at a paper speed of 63.5 cm/s. The output of the potentiometer was calibrated to the center and upper and lower boundaries of the target area prior to each testing session. E M G activity was monitored by two pairs of surface electrodes (Beckman) positioned parallel to the long axis of the bicep (agonist) and two electrode pairs over the tricep (antagonist) muscles. The interelectrode distance was approx. 3.0 cm, and interelectrode resistance less than 10 kQ for each electrode pair. A ground electrode was placed over the right acromion process. Four channels of raw E M G signals were processed using 4 Beckman couplers (Type 9852A) in combination with a Beckman R-411 6-channel recorder. Amplifiers were calibrated prior to each testing session, and the high frequency response set at maximum. Analog signals were displayed using curvilinear pens and a paper speed of 100 mm/s. Blindfolded subjects were required to complete a series of training trials until the response was 'learned'. This was achieved when the subject completed 5 consecutive trials with each trial terminated in the target area in less than 200 ms. To facilitate acquisition of the skill knowledge of results about the magnitude and direction of the response error (e.g. +1.5 U) and the speed of the movement was provided after each trial. All training trials were completed with a 1.1 kg control spring load attached to the slide. After the 5th successful training trial, one more movement was completed and designated a test load trial. Time-displacement and E M G data were recorded during each test load trial. This sequence of training trials followed by a single test load trial was repeated 8 times by each subject. For each test load trial, either the 1.1 kg control spring load was removed and then reapplied to the slide or was replaced by a 1.1, 1.5 or 2.0 kg inertial test load. The procedures for interchanging the spring and inertial loads were identical for each test load trial. The spring load was randomly applied on 4 of the test load trials, and one of the inertial loads applied on the other 4 test load trials. Four subjects were ran-
domly assigned to one of 3 independent groups. All groups trained with the spring load, but differed in the size of the inertial load. Groups 1, 2 and 3 represent inertial test loadings of 1.1, 1.5 and 2.0 kg, respectively. Each E M G record was evaluated by two raters on two separate occasions (at least 24 h apart), and classified as being either coactivation or triphasic in nature. The criteria used for classification were as follows. A triphasic pattern was characterized by two discrete bursts of agonist activity separated by a silent period, during which a distinct pulse of antagonist activity was observed. A typical triphasic pattern observed during spring load conditions is presented in Fig. 1A. When a single burst of agonist activity occurred simultaneously with a sustained and relatively constant level of innervation of the antagonist muscle, it was considered to be characteristic of a coactivation pattern. An example of this E M G pattern observed when the movement was unexpectedly loaded by an inertial load is illustrated in Fig. lB. Inter-rater reliability for classifying'EMG patterns during spring load trials and inertial load trials was Ca = 0.609 and 0.483, respectively. These reliability coefficients were transformed into ~ statistics 5 and found to be significant, ~ (1) = 17.8 and 11.2, P < 0.01. For spring load trials, 69% of the E M G patterns were classified as triphasic patterns, whereas on inertial load trials 94.0% were classified as coactivation patterns. There were no differences in frequency or type of E M G patterns as a function of load (i.e. groups 1,2, and 3). The data of Experiment 1 support the predicted relationship between the type of external load force and the associated E M G activity. Practising the movement with an elastic spring load attached to the apparatus presumably biased the CNS to preset the musculature to generate a triphasic activation pattern. However, when the response was unexpectedly perturbed by an inertial load, the results indicated that the CNS reset the neuromuscular activation pattern prior to movement termination to produce an optimal control pattern of coactivation. A second experiment was conducted to determine if the principle of modifying the neuromuscular innervation pattern observed in the first experiment occurs when the sequence of loading is reversed (i.e.
137 inertial
to spring load). The procedures
periment
2 were identical
used in Ex-
to those adopted in the first
experiment with one exception. Twelve subjects not used in Experiment 1, completed the training trials with a 1.1 kg inertial load attached to the apparatus. On test load trials, the inertial load was detached and then reattached to the apparatus or was replaced by a 1.1,1.5or2.0kgspringload.
Each EMG record for the test load trials was evaluated using procedures
and waveform
criteria identi-
cal to those used in Experiment 1. Typical coactivation and triphasic patterns produced by subjects during the second experiment are presented in Fig. 2A,B. Inter-rater reliability for classifying EMG patterns for the inertial 0.342
and
0.295,
and spring load trials was q = respectively.
These
coefficients
0
A
AVG. DISP
in mVOLTS
AVG. DISP.
0.4375,
in mVOLTS
0.4375,
0.0625 t
EMG
EMG
UNITS
I
0.51
0.375 0.25
0.75 ‘I-
1 I
illl
-0.25
I
0.S
u
0 -0.125
UNITS
I
II
0.125
/
0.2s 0 -0.25 -0.5
-0.5 r -0.375
-Dd 0 EMG
UNITS
EMG
I
0.375
I
1
400
500
-0.25L
1.51
1 100
1 200
I 300
I 400
500
I 100
1 200
I 300
1 400
I 500
UNITS
-1.5 -2-2.5 D
L
1
100
200 TIME
, 300 (mSEC
1
0
TIME
(mSEC)
Fig. 1. Average time-displacement and EMG activation patterns associated with rapid arm movements perturbed by a control spring load (A) or a test inertial load (B). The top trace, displacement; middle trace, agonist EMG record; lower trace, antagonist EMG record. All 3 records are the average of 4 single records for a subject in Group 2. The spring load was 1.0 kg and the test inertial load was 2.0 kg. Average displacement is marked in mV with 1.0 mV representing 2.13 rad. The two horizontal lines in the displacement graph represent the target area. The upper and lower traces are time-locked to the onset of agonist activity and the time base is expressed as beginning 50 ms prior to this point. The difference between the records in A and B is most noticeable in the activity of the antagonist. In A a single discrete antagonist burst is consistent with a triphasic pattern. In B antagonist activity is sustained throughout the last half of the movement and indicates a coactivation pattern.
138 were statistically significant, Z: (1) = 5.61 and 4.12, P < 0.05. For inertial test load trials, 83% of the E M G patterns were observed to be coactivation and 79% of the spring test load trials were classified as triphasic. The limb-apparatus system has a total stiffness component equal to the sum of the stiffness inherent in both the limb and the apparatus. Thus, to produce an efficient and accurate m o v e m e n t to a target, the limb stiffness must be scaled to the mechanical stiffness of the apparatus. An error in scaling will result in either an underdamped or overdamped system,
which in turn will produce inaccurate positional control and significant alteration in m o v e m e n t velocity and acceleration. In Experiment 1, the relatively large stiffness component provided by the spring load during the testload trials reduced the need for limb-generated stiffness and motor control was maintained using a more energy-efficient triphasic pattern. H o w e v e r , during unexpected inertial loading, this relatively low level of limb stiffness generated in anticipation of the spring load represented an underdamped response system, which if left unchanged would have resulted
AVG. DISP. in m V O L T S
AVG. DISP. in reVOLTS
0.4375
0,437~.
f
0.375
0.375 0.3125
0.3125 0.25
0.25
0.1875
0.1875
0.125
0.125
0.0625
0.0625 I
100
0
200
300
0
400
500 EM(
UNITS
EM,
0
0,75
(175
0.5
0.5
0.25
0.25 C
0.25
-0.25
-0.5
-0.5
0 EMG
-0.75
I
1
I
t
100
200
300
400
500
0.5
1
0.25
0.5
e
O
-o.25
-0.5
-o.5
0
0
-o.75
I
1
I
100
200 TIME
300 (rnSEC)
400
500
300
400
500
lOO
200
3oo
400
500
I
I
200
300
400
UNITS
EM(
NITS
1.5
200
UNITS
C
-0.7 ~_
100
I
o
100
500
TIME (mSEC)
Fig. 2. Average time-displacement and EMG activation patterns associated with rapid arm movements perturbed by a control inertial load (A) and a test spring load (B). The records are averages for a subject in Group 1. Inertial load was 1.0 kg and the springload 1.0 kg. Tracings of markings for the graphs are the same as in Fig. 1. Under inertial load conditions (A), the antagonist indicates coactivation while during unexpected loading with a spring load the antagonist conforms to a triphasic activation pattern,
139 in an overestimation of the target. To achieve the desired positional accuracy, limb stiffness needed to be increased which, as predicted, was accomplished by the CNS resetting the neuromuscular activation pattern from a triphasic pattern to one of coactivation. In Experiment 2, the switch from an inertial to spring load required a decrease in limb stiffness. Failure to effect this change would have resulted in a severely overdamped response causing degradation of movement efficiency and accuracy. In this case, modification of limb stiffness was observed as a switch from a coactivation pattern to a triphasic pattern. Therefore, the results of Experiment 2 fully complemented the findings of the first experiment, and collectively the two experiments demonstrated that regardless of the sequence of loading, the CNS can modify the muscular innervation patterns prior to completion of a response. Another significant feature of the data was that increases in the size of the test load did not affect modification of the neuromuscular activation pattern. It has been previously demonstrated that increased loading does produce different levels of neuromuscular innervation 6, but according to the present results, amplitude modulation does not alter the overall configuration of the E M G waveform specified by the CNS. An interesting aspect to the present results concerns the nature of the signal that triggers the CNS to change the activation pattern from one type to another. A common explanation of rapid error correction during visually guided responses is provided by the concept of efference copy 17 or corollary discharge 12. Accordingly, error detection-error correction is thought to occur at a central level through the use of an 'internal' feedback loop. Response-produced feedback need not be included as part of this internal loop meaning that errors can be detected and corrected prior to movement initiation. The notion of modifying activation patterns through a mechanism such as efference copy or corollary discharge is not applicable to the unsighted and unexpectedly loaded movements used in the present experiments. Based on the assu/nption that the conditions of the training trials are constant, and in the absence of visual input to indicate that conditions have changed, the subject will generate a set of efferent commands founded on the expectation that the
mechanical interaction between the limb and load will be the same on the upcoming trial as it was for past training trials. Furthermore, the same set of efferent commands will be discharged to the musculature unless a difference in load is detected. However, detection of any differences in the stiffness properties of each load is only possible after movement initiation, which eliminates the use of a central system correcting errors prior to movement initiation. In the absence of prior information and visual input it seems reasonable to assume that changes in the mechanical interaction between limb and apparatus are detected by the proprioceptive system. This leads to the possibility that peripheral input drives the central structures to regulate limb stiffness in accordance with any detected alteration in external load conditions. The limitation with proposing a peripheralcentral interaction is to explain how such a feedback loop can work during rapid movements (i.e. movements completed in <200 ms). During these fast responses the transmission delays associated with information traversing the transcortical feedback loop often exceed the total movement time, so that no corrections were made to the response observed prior to movement termination. In Experiment 1, where the unexpected application of the inertial load increased movement time to 307.8 ms, sufficient time was available for this loop to be completed and correct modification of limb stiffness to be made. In Experiment 2, however, the unexpected imposition of the spring load resulted in movement time decreasing to 158.6 ms, which placed severe temporal limitations on the time available to complete the feedback loop. Since the regulation of limb stiffness cannot occur prior to movement initiation through a central mechanism such as efference copy, but must be completed fast enough to allow sufficient time for the new efferent commands to be enacted prior to movement termination, the peripherally generated signal to switch activation patterns must occur during a narrow window of time corresponding to the earliest part of the movement. A follow-up examination of the time of onset and time of maximal innervation for the agonist and antagonist muscles revealed that although innervation of the antagonist occurred 15 ms prior to movement initiation, the agonist was fully involved during the early part of the movement. The agonist was acti-
14() vated 83.8 ms prior to movement initiation and was maximally innervated at the point of m o v e m e n t onset. This involvement can be seen in Fig. 2, which also indicates that the major contribution of the an-
used visually guided movements, which as previously noted, represent a different class of responses. This leaves the contribution of the agonist pulse to the control of rapid, unsighted, isotonic movements with
tagonist does not occur until later in the response.
variable limb stiffness requirements virtually unknown.
These data and observations suggest that peripheral information generated primarily as a result of agonist activity during the first part of the m o v e m e n t
In summary, data are presented which demonstrate that limb stiffness associated with a rapid un-
is responsible for signaling the CNS to reset limb stiffness when a new load is encountered. This is not an
sighted response can be regulated prior to movement termination. Regulation was imposed through the
entirely new idea, having been previously proposed by Allum j. He suggested that during isometric con-
CNS resetting current neuromuscular activation to a pattern more conductive to producing the required
tractions against preexisting force levels, the agonist
limb stiffness. Furthermore, the signal to reset limb stiffness when an unexpected load type was encoun-
is used as a medium latency E M G response which provides a pulse-test signal about external loading conditions to the CNS. Based on the outcome of the pulse test, the muscles are activated to cater to the external load conditions. Other investigators have
tered appears to be composed of peripherally generated sensory information resulting from a high level of agonist activity occurring during the early part of the movement.
also studied the role of the agonist burst, but have
1 Allum, J.H.J., Responses to load disturbances in human shoulder muscles: the hypothesis that one component is a pulse test information signal, Exp. Brain Res., 22 (1975) 307-326. 2 Bawa, P.N.S. and Dickinson, J., Force as'the controlling muscle variable in limb movement, Behav. Brain Sci., 5 (1982) 543-544. 3 Bizzi, E., Dev, P., Morasso, P. and Polit, A., Effects of load disturbance during centrally initiated movements, J. Neurophysiol., 41 (1978)542-566. 4 Brown, S.H.C. and Cooke, J.D., Responses to force perturbations preceding voluntary human arm movements, Brain Research, 22 (1981) 350-355. 5 Guildford, J.P., Psychometric Methods, McGraw Hill, New York, 1954. 6 Hannaford, B., Lakshminarayanan, V., Stark, L. and Nam, M., Electromyographic evidence of neurological controller signals with viscous load, J. Mot. Behav., 16 (1984) 255-274. 7 Hoffer, J.A., Central control and reflex regulation of mcchanical impedance: the basis for a unified motor-control scheme, Behav. Brain Sci., 5 (1982) 548-549. 8 Hogan, N., Adaptive control of mechanical impedance by coactivation of antagonist muscles, IEEE Trans. Eng., 29 (1984) 681-690. 9 Houk, J.C., Regulation of stiffness by skeletomotor reflexes, Annu. Rev. Physiol., 41 (1979)99-114. 10 Lestienne, F., Effects of inertial load and velocity on the
braking process of voluntary limb movements, Exp. Brain Res., 35 (1979) 407-418. 11 Schmidt, R.A. and McGown, C.M., Terminal accuracy of unexpectedly loaded rapid movements: evidence for a mass-spring mechanism in programming, J. Mot. Behav., 12 (1980) 149-162. 12 Sperry, R.W., Neural basis of the spontaneous optokinetic response produced by visual inversion, J. Comp. Physiol. Psychol., 43 (1950) 482-489. 13 Simmons, R.W. and Richardson, C., Maintenance of equilibrium point control during an unexpectedly loaded rapid limb movement, Brain Research, 302 (1984) 239-244. 14 Stark, L., Neurological ballistic movements: sampled data or intermittent open-loop control, Behav, Brain Sci., 5 (1982) 564-566. 15 Stein, R.B., What muscle variable(s) does the nervous system control in limb movements? Behav. Brain Sci.. 5 (1982) 535-577. 16 Terzuolo, C.A., Soechting, J.F. and Palminteri, R., Studies on the control of some simple motor tasks. III. Comparison of the EMG pattern during ballistically initiated movements in man and squirrel monkey, Brain Research, 62 (1973) 242-246. 17 Van Hoist, E. and Minelstaedt, H., The reafference principle. Interaction between the central nervous system and the periphery, 1950. In Selected Papers of Erich yon Holst, The Behavioural Physiology of Animals and Man, Methuen, London, 1973.